Exemple #1
0
 public TumblrLikedByCrawler(IShellService shellService, CancellationToken ct, PauseToken pt,
                             IProgress <DownloadProgress> progress, ICrawlerService crawlerService, IWebRequestFactory webRequestFactory, ISharedCookieService cookieService, IDownloader downloader, BlockingCollection <TumblrPost> producerConsumerCollection, IBlog blog)
     : base(shellService, ct, progress, webRequestFactory, cookieService, producerConsumerCollection, blog)
 {
     this.downloader = downloader;
     this.pt         = pt;
 }
Exemple #2
0
        public ICrawler GetCrawler(IBlog blog, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, IShellService shellService, ICrawlerService crawlerService, IManagerService managerService)
        {
            IPostQueue <TumblrPost> postQueue    = GetProducerConsumerCollection();
            IFiles             files             = LoadFiles(blog, managerService);
            IWebRequestFactory webRequestFactory = GetWebRequestFactory();
            IImgurParser       imgurParser       = GetImgurParser(webRequestFactory, ct);
            IGfycatParser      gfycatParser      = GetGfycatParser(webRequestFactory, ct);

            switch (blog.BlogType)
            {
            case BlogTypes.tumblr:
                IPostQueue <TumblrCrawlerData <DataModels.TumblrApiJson.Post> > jsonApiQueue = GetJsonQueue <DataModels.TumblrApiJson.Post>();
                return(new TumblrBlogCrawler(shellService, ct, pt, progress, crawlerService, webRequestFactory, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, managerService, blog, files, postQueue), GetTumblrJsonDownloader(shellService, ct, pt, jsonApiQueue, crawlerService, blog), GetTumblrApiJsonToTextParser(blog), imgurParser, gfycatParser, GetWebmshareParser(), postQueue, jsonApiQueue, blog));

            case BlogTypes.tmblrpriv:
                IPostQueue <TumblrCrawlerData <DataModels.TumblrSvcJson.Post> > jsonSvcQueue = GetJsonQueue <DataModels.TumblrSvcJson.Post>();
                return(new TumblrHiddenCrawler(shellService, ct, pt, progress, crawlerService, webRequestFactory, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, managerService, blog, files, postQueue), GetTumblrJsonDownloader(shellService, ct, pt, jsonSvcQueue, crawlerService, blog), GetTumblrSvcJsonToTextParser(blog), imgurParser, gfycatParser, GetWebmshareParser(), postQueue, jsonSvcQueue, blog));

            case BlogTypes.tlb:
                return(new TumblrLikedByCrawler(shellService, ct, pt, progress, crawlerService, webRequestFactory, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, managerService, blog, files, postQueue), postQueue, blog));

            case BlogTypes.tumblrsearch:
                return(new TumblrSearchCrawler(shellService, ct, pt, progress, crawlerService, webRequestFactory, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, managerService, blog, files, postQueue), postQueue, blog));

            case BlogTypes.tumblrtagsearch:
                return(new TumblrTagSearchCrawler(shellService, ct, pt, progress, crawlerService, webRequestFactory, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, managerService, blog, files, postQueue), postQueue, blog));

            default:
                throw new ArgumentException("Website is not supported!", "blogType");
            }
        }
        public ICrawler GetCrawler(BlogTypes blogtype, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, IShellService shellService, ICrawlerService crawlerService, IBlog blog)
        {
            BlockingCollection <TumblrPost> producerConsumerCollection = GetProducerConsumerCollection();
            IFiles files = LoadFiles(blog);

            switch (blogtype)
            {
            case BlogTypes.tumblr:
                return(new TumblrBlogCrawler(shellService, ct, pt, progress, crawlerService, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, blog, files, producerConsumerCollection), producerConsumerCollection, blog));

            case BlogTypes.tmblrpriv:
                return(new TumblrPrivateCrawler(shellService, ct, pt, progress, crawlerService, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, blog, files, producerConsumerCollection), producerConsumerCollection, blog));

            case BlogTypes.tlb:
                return(new TumblrLikedByCrawler(shellService, ct, pt, progress, crawlerService, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, blog, files, producerConsumerCollection), producerConsumerCollection, blog));

            case BlogTypes.tumblrsearch:
                return(new TumblrSearchCrawler(shellService, ct, pt, progress, crawlerService, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, blog, files, producerConsumerCollection), producerConsumerCollection, blog));

            case BlogTypes.tumblrtagsearch:
                return(new TumblrTagSearchCrawler(shellService, ct, pt, progress, crawlerService, cookieService, GetTumblrDownloader(ct, pt, progress, shellService, crawlerService, blog, files, producerConsumerCollection), producerConsumerCollection, blog));

            default:
                throw new ArgumentException("Website is not supported!", "blogType");
            }
        }
Exemple #4
0
 public TumblrSearchCrawler(IShellService shellService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress,
                            ICrawlerService crawlerService, IWebRequestFactory webRequestFactory, ISharedCookieService cookieService,
                            IDownloader downloader, IPostQueue <TumblrPost> postQueue, IBlog blog)
     : base(shellService, crawlerService, ct, progress, webRequestFactory, cookieService, postQueue, blog)
 {
     this.downloader = downloader;
     this.pt         = pt;
 }
Exemple #5
0
 public TumblrBlogCrawler(IShellService shellService, CancellationToken ct, PauseToken pt,
                          IProgress <DownloadProgress> progress, ICrawlerService crawlerService, IWebRequestFactory webRequestFactory, ISharedCookieService cookieService, IDownloader downloader, IImgurParser imgurParser, IGfycatParser gfycatParser, IWebmshareParser webmshareParser, BlockingCollection <TumblrPost> producerConsumerCollection, IBlog blog)
     : base(shellService, ct, pt, progress, crawlerService, webRequestFactory, cookieService, downloader, producerConsumerCollection, blog)
 {
     this.imgurParser     = imgurParser;
     this.gfycatParser    = gfycatParser;
     this.webmshareParser = webmshareParser;
 }
Exemple #6
0
 public TumblrJsonDownloader(IShellService shellService, CancellationToken ct, PauseToken pt, IPostQueue <TumblrCrawlerData <T> > jsonQueue, ICrawlerService crawlerService, IBlog blog)
 {
     this.shellService   = shellService;
     this.crawlerService = crawlerService;
     this.blog           = blog;
     this.ct             = ct;
     this.pt             = pt;
     this.jsonQueue      = jsonQueue;
 }
 public TumblrXmlDownloader(IShellService shellService, CancellationToken ct, PauseToken pt, IPostQueue <TumblrCrawlerData <XDocument> > xmlQueue, ICrawlerService crawlerService, IBlog blog)
 {
     this.shellService   = shellService;
     this.crawlerService = crawlerService;
     this.blog           = blog;
     this.ct             = ct;
     this.pt             = pt;
     this.xmlQueue       = xmlQueue;
 }
 public TumblrXmlDownloader(IShellService shellService, PauseToken pt, IPostQueue <CrawlerData <XDocument> > xmlQueue, ICrawlerService crawlerService, IBlog blog, CancellationToken ct)
 {
     _shellService   = shellService;
     _crawlerService = crawlerService;
     _blog           = blog;
     _ct             = ct;
     _pt             = pt;
     _xmlQueue       = xmlQueue;
 }
Exemple #9
0
 public TumblrXmlDownloader(IShellService shellService, CancellationToken ct, PauseToken pt, BlockingCollection <TumblrCrawlerXmlData> xmlQueue, ICrawlerService crawlerService, IBlog blog)
 {
     this.shellService   = shellService;
     this.crawlerService = crawlerService;
     this.blog           = blog;
     this.ct             = ct;
     this.pt             = pt;
     this.xmlQueue       = xmlQueue;
 }
Exemple #10
0
        public static async Task SomeMethodAsync(PauseToken pause)
        {
            await Task.Delay(500);

            Console.WriteLine("Before await pause.WaitWhilePausedAsync()");
            await pause.WaitWhilePausedWithResponseAsyc();

            Console.WriteLine("After await pause.WaitWhilePausedAsync()");
        }
        public static async Task SomeMethodAsync(PauseToken pause)
        {
            for (int i = 0; i < 100; i++)
            {
                Console.WriteLine(i);
                await Task.Delay(100);

                await pause.WaitWhilePausedAsync();
            }
        }
 protected AbstractCrawler(IShellService shellService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, ICrawlerService crawlerService, IDownloader downloader, BlockingCollection <TumblrPost> producerConsumerCollection, IBlog blog)
 {
     this.shellService               = shellService;
     this.crawlerService             = crawlerService;
     this.downloader                 = downloader;
     this.producerConsumerCollection = producerConsumerCollection;
     this.blog     = blog;
     this.ct       = ct;
     this.pt       = pt;
     this.progress = progress;
 }
Exemple #13
0
        private async Task SearchImages(string url, CancellationToken cancelToken, PauseToken pauseToken, string folder = null)
        {
            string[] links;

            if (!(url.StartsWith("http://", StringComparison.OrdinalIgnoreCase) ||
                  url.StartsWith("https://", StringComparison.OrdinalIgnoreCase))
                )
            {
                url = "http://" + url;
            }

            this.txtUrl.Text = url;
            int attempts = 0;

            while (true)
            {
                await pauseToken.WaitWhilePausedAsync();

                if (m_cancel)
                {
                    return;
                }

                links = await m_downloader.GetImageLinks(url);

                if (links.Length == 0 && m_downloader.HasErrors)
                {
                    this.ShowLastErrors();
                    if (++attempts == 10)
                    {
                        break;
                    }
                }
                else
                {
                    break;
                }
            }

            m_links = links;

            this.clbImageList.BeginUpdate();
            this.clbImageList.Items.Clear();
            this.clbImageList.Items.AddRange(m_links);
            this.clbImageList.EndUpdate();
            this.tssItemCount.Text = string.Format("Images found: {0}", this.clbImageList.Items.Count);

            this.ShowLastErrors();

            if (m_links.Length > 0)
            {
                await this.DownloadImages(cancelToken, pauseToken, folder);
            }
        }
 public TwitterCrawler(IShellService shellService, ICrawlerService crawlerService, IProgress <DownloadProgress> progress, IWebRequestFactory webRequestFactory,
                       ISharedCookieService cookieService, IPostQueue <AbstractPost> postQueue, IPostQueue <CrawlerData <Tweet> > jsonQueue, IBlog blog, IDownloader downloader,
                       ICrawlerDataDownloader crawlerDataDownloader, PauseToken pt, CancellationToken ct)
     : base(shellService, crawlerService, progress, webRequestFactory, cookieService, postQueue, blog, downloader, pt, ct)
 {
     this.downloader = downloader;
     this.downloader.ChangeCancellationToken(Ct);
     this.jsonQueue             = jsonQueue;
     this.crawlerDataDownloader = crawlerDataDownloader;
     this.crawlerDataDownloader.ChangeCancellationToken(Ct);
 }
Exemple #15
0
        public static async Task SomeMethodAsync(PauseToken pause)
        {
            while (true)
            {
                await Task.Delay(1000).ConfigureAwait(false);

                Console.WriteLine("Before await pause.WaitWhilePausedAsync()");
                await pause.WaitWhilePausedWithResponseAsyc();

                Console.WriteLine("After await pause.WaitWhilePausedAsync()");
            }
        }
Exemple #16
0
 protected AbstractDownloader(IShellService shellService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, BlockingCollection <TumblrPost> producerConsumerCollection, FileDownloader fileDownloader, ICrawlerService crawlerService = null, IBlog blog = null, IFiles files = null)
 {
     this.shellService               = shellService;
     this.crawlerService             = crawlerService;
     this.blog                       = blog;
     this.files                      = files;
     this.ct                         = ct;
     this.pt                         = pt;
     this.progress                   = progress;
     this.producerConsumerCollection = producerConsumerCollection;
     this.fileDownloader             = fileDownloader;
 }
Exemple #17
0
 public TumblrSearchCrawler(IShellService shellService, CancellationToken ct, PauseToken pt,
                            IProgress <DownloadProgress> progress, ICrawlerService crawlerService, IWebRequestFactory webRequestFactory,
                            ISharedCookieService cookieService, IDownloader downloader, ITumblrParser tumblrParser, IImgurParser imgurParser,
                            IGfycatParser gfycatParser, IWebmshareParser webmshareParser, IMixtapeParser mixtapeParser, IUguuParser uguuParser,
                            ISafeMoeParser safemoeParser, ILoliSafeParser lolisafeParser, ICatBoxParser catboxParser,
                            IPostQueue <TumblrPost> postQueue, IBlog blog)
     : base(shellService, crawlerService, ct, pt, progress, webRequestFactory, cookieService, tumblrParser, imgurParser,
            gfycatParser, webmshareParser, mixtapeParser, uguuParser, safemoeParser, lolisafeParser, catboxParser, postQueue,
            blog)
 {
     this.downloader = downloader;
 }
Exemple #18
0
        private async Task RunCrawlerTasksAsync(PauseToken pt, CancellationToken ct)
        {
            while (true)
            {
                if (ct.IsCancellationRequested)
                {
                    break;
                }

                if (pt.IsPaused)
                {
                    pt.WaitWhilePausedWithResponseAsyc().Wait();
                }

                Monitor.Enter(_lockObject);
                if (_crawlerService.ActiveItems.Count < QueueManager.Items.Count)
                {
                    IEnumerable <QueueListItem> queueList = QueueManager.Items.Except(_crawlerService.ActiveItems);
                    QueueListItem nextQueueItem           = queueList.First();
                    IBlog         blog = nextQueueItem.Blog;

                    ICrawler crawler = _crawlerFactory.GetCrawler(blog, new Progress <DownloadProgress>(), pt, ct);
                    crawler.IsBlogOnlineAsync().Wait(4000);
                    crawler.Dispose();

                    if (_crawlerService.ActiveItems.Any(item =>
                                                        item.Blog.Name.Equals(nextQueueItem.Blog.Name) &&
                                                        item.Blog.BlogType.Equals(nextQueueItem.Blog.BlogType)))
                    {
                        QueueOnDispatcher.CheckBeginInvokeOnUI(() => QueueManager.RemoveItem(nextQueueItem));
                        Monitor.Exit(_lockObject);
                        continue;
                    }

                    if (!nextQueueItem.Blog.Online)
                    {
                        QueueOnDispatcher.CheckBeginInvokeOnUI(() => QueueManager.RemoveItem(nextQueueItem));
                        Monitor.Exit(_lockObject);
                        continue;
                    }

                    _crawlerService.AddActiveItems(nextQueueItem);
                    Monitor.Exit(_lockObject);
                    await StartSiteSpecificDownloaderAsync(nextQueueItem, pt, ct);
                }
                else
                {
                    Monitor.Exit(_lockObject);
                    await Task.Delay(4000, ct);
                }
            }
        }
Exemple #19
0
 protected AbstractDownloader(IShellService shellService, IManagerService managerService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, IPostQueue <TumblrPost> postQueue, FileDownloader fileDownloader, ICrawlerService crawlerService = null, IBlog blog = null, IFiles files = null)
 {
     this.shellService   = shellService;
     this.crawlerService = crawlerService;
     this.managerService = managerService;
     this.blog           = blog;
     this.files          = files;
     this.ct             = ct;
     this.pt             = pt;
     this.progress       = progress;
     this.postQueue      = postQueue;
     this.fileDownloader = fileDownloader;
 }
Exemple #20
0
        /// <summary>
        /// Holds the thread while the pause token is paused, if cancellation is detected this will return true (false otherwise)
        /// </summary>
        /// <param name="cancellationToken"></param>
        /// <param name="pauseToken"></param>
        /// <returns></returns>
        private bool PauseHolder(CancellationToken cancellationToken, PauseToken pauseToken)
        {
            while (pauseToken.IsPaused)
            {
                if (cancellationToken.IsCancellationRequested)
                {
                    return(true);
                }
                Thread.Sleep(10);
            }

            return(false);
        }
Exemple #21
0
        public GameFolder([NotNull] DirectoryInfo directory, PauseToken pauseToken)
        {
            DirectoryInfo = directory;
            PauseToken    = pauseToken;
            IsJunction    = JunctionPoint.Exists(directory);
            if (IsJunction)
            {
                Size           = JUNCTION_POINT_SIZE;
                JunctionTarget = JunctionPoint.GetTarget(directory);
            }

            UpdatePropertiesFromSubdirectoriesAsync().Forget();
        }
Exemple #22
0
 public TumblrBlogCrawler(IShellService shellService, ICrawlerService crawlerService, IWebRequestFactory webRequestFactory,
                          ISharedCookieService cookieService, IDownloader downloader, ICrawlerDataDownloader crawlerDataDownloader,
                          ITumblrToTextParser <Post> tumblrJsonParser, ITumblrParser tumblrParser, IImgurParser imgurParser,
                          IGfycatParser gfycatParser, IWebmshareParser webmshareParser, IUguuParser uguuParser, ICatBoxParser catboxParser,
                          IPostQueue <AbstractPost> postQueue, IPostQueue <CrawlerData <Post> > jsonQueue, IBlog blog,
                          IProgress <DownloadProgress> progress, PauseToken pt, CancellationToken ct)
     : base(shellService, crawlerService, webRequestFactory, cookieService, tumblrParser, imgurParser, gfycatParser,
            webmshareParser, uguuParser, catboxParser, postQueue, blog, downloader, crawlerDataDownloader, progress, pt, ct)
 {
     this.downloader = downloader;
     this.downloader.ChangeCancellationToken(Ct);
     this.tumblrJsonParser = tumblrJsonParser;
     this.jsonQueue        = jsonQueue;
 }
Exemple #23
0
 protected AbstractCrawler(IShellService shellService, ICrawlerService crawlerService, CancellationToken ct, PauseToken pt,
                           IProgress <DownloadProgress> progress, IWebRequestFactory webRequestFactory, ISharedCookieService cookieService,
                           IPostQueue <TumblrPost> postQueue, IBlog blog)
 {
     this.shellService      = shellService;
     this.crawlerService    = crawlerService;
     this.webRequestFactory = webRequestFactory;
     this.cookieService     = cookieService;
     this.postQueue         = postQueue;
     this.blog     = blog;
     this.pt       = pt;
     this.ct       = ct;
     this.progress = progress;
 }
Exemple #24
0
 protected AbstractCrawler(IShellService shellService, ICrawlerService crawlerService, IProgress <DownloadProgress> progress, IWebRequestFactory webRequestFactory,
                           ISharedCookieService cookieService, IPostQueue <TumblrPost> postQueue, IBlog blog,
                           PauseToken pt, CancellationToken ct)
 {
     ShellService      = shellService;
     CrawlerService    = crawlerService;
     WebRequestFactory = webRequestFactory;
     CookieService     = cookieService;
     PostQueue         = postQueue;
     Blog     = blog;
     Progress = progress;
     Pt       = pt;
     Ct       = ct;
 }
 protected AbstractDownloader(IShellService shellService, IManagerService managerService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, IPostQueue <TumblrPost> postQueue, FileDownloader fileDownloader, ICrawlerService crawlerService = null, IBlog blog = null, IFiles files = null)
 {
     this.shellService   = shellService;
     this.crawlerService = crawlerService;
     this.managerService = managerService;
     this.blog           = blog;
     this.files          = files;
     this.ct             = ct;
     this.pt             = pt;
     this.progress       = progress;
     this.postQueue      = postQueue;
     this.fileDownloader = fileDownloader;
     _saveTimer          = new Timer(_ => OnSaveTimedEvent(), null, SAVE_TIMESPAN_SECS * 1000, SAVE_TIMESPAN_SECS * 1000);
 }
 protected AbstractTumblrCrawler(IShellService shellService, ICrawlerService crawlerService, IWebRequestFactory webRequestFactory, ISharedCookieService cookieService,
                                 ITumblrParser tumblrParser, IImgurParser imgurParser, IGfycatParser gfycatParser, IWebmshareParser webmshareParser, IUguuParser uguuParser,
                                 ICatBoxParser catboxParser, IPostQueue <AbstractPost> postQueue, IBlog blog, IDownloader downloader, ICrawlerDataDownloader crawlerDataDownloader,
                                 IProgress <DownloadProgress> progress, PauseToken pt, CancellationToken ct)
     : base(shellService, crawlerService, progress, webRequestFactory, cookieService, postQueue, blog, downloader, pt, ct)
 {
     this.crawlerDataDownloader = crawlerDataDownloader;
     this.crawlerDataDownloader?.ChangeCancellationToken(Ct);
     this.TumblrParser    = tumblrParser;
     this.ImgurParser     = imgurParser;
     this.GfycatParser    = gfycatParser;
     this.WebmshareParser = webmshareParser;
     this.UguuParser      = uguuParser;
     this.CatboxParser    = catboxParser;
 }
 public TumblrBlogCrawler(IShellService shellService, ICrawlerService crawlerService, IWebRequestFactory webRequestFactory,
                          ISharedCookieService cookieService, IDownloader downloader, ICrawlerDataDownloader crawlerDataDownloader,
                          ITumblrToTextParser <Post> tumblrJsonParser, ITumblrParser tumblrParser, IImgurParser imgurParser,
                          IGfycatParser gfycatParser, IWebmshareParser webmshareParser, IMixtapeParser mixtapeParser,
                          IUguuParser uguuParser, ISafeMoeParser safemoeParser, ILoliSafeParser lolisafeParser, ICatBoxParser catboxParser,
                          IPostQueue <TumblrPost> postQueue, IPostQueue <TumblrCrawlerData <Post> > jsonQueue, IBlog blog,
                          IProgress <DownloadProgress> progress, PauseToken pt, CancellationToken ct)
     : base(shellService, crawlerService, webRequestFactory, cookieService, tumblrParser, imgurParser, gfycatParser,
            webmshareParser, mixtapeParser, uguuParser, safemoeParser, lolisafeParser, catboxParser, postQueue, blog, progress, pt,
            ct)
 {
     this.downloader            = downloader;
     this.tumblrJsonParser      = tumblrJsonParser;
     this.jsonQueue             = jsonQueue;
     this.crawlerDataDownloader = crawlerDataDownloader;
 }
Exemple #28
0
        protected AbstractDownloader(IShellService shellService, IManagerService managerService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress, IPostQueue <AbstractPost> postQueue, FileDownloader fileDownloader, ICrawlerService crawlerService = null, IBlog blog = null, IFiles files = null)
        {
            this.shellService   = shellService;
            this.crawlerService = crawlerService;
            this.managerService = managerService;
            this.blog           = blog;
            this.files          = files;
            this.ct             = ct;
            this.pt             = pt;
            this.progress       = progress;
            this.postQueue      = postQueue;
            this.fileDownloader = fileDownloader;
            Progress <Exception> prog = new Progress <Exception>((e) => shellService.ShowError(e, Resources.CouldNotSaveBlog, blog.Name));

            _saveTimer = new Timer(_ => OnSaveTimedEvent(prog), null, SAVE_TIMESPAN_SECS * 1000, SAVE_TIMESPAN_SECS * 1000);
        }
 public ICrawler GetCrawler(IBlog blog, IProgress<DownloadProgress> progress, PauseToken pt, CancellationToken ct)
 {
     IPostQueue<TumblrPost> postQueue = GetProducerConsumerCollection();
     IFiles files = LoadFiles(blog);
     IWebRequestFactory webRequestFactory = GetWebRequestFactory();
     IImgurParser imgurParser = GetImgurParser(webRequestFactory, ct);
     IGfycatParser gfycatParser = GetGfycatParser(webRequestFactory, ct);
     switch (blog.BlogType)
     {
         case BlogTypes.tumblr:
             IPostQueue<TumblrCrawlerData<Post>> jsonApiQueue = GetJsonQueue<Post>();
             return new TumblrBlogCrawler(shellService, crawlerService, webRequestFactory, cookieService,
                 GetTumblrDownloader(progress, blog, files, postQueue, pt, ct), GetTumblrJsonDownloader(jsonApiQueue, blog, pt, ct),
                 GetTumblrApiJsonToTextParser(blog), GetTumblrParser(), imgurParser, gfycatParser, GetWebmshareParser(),
                 GetMixtapeParser(), GetUguuParser(), GetSafeMoeParser(), GetLoliSafeParser(), GetCatBoxParser(), postQueue,
                 jsonApiQueue, blog, progress, pt, ct);
         case BlogTypes.tmblrpriv:
             IPostQueue<TumblrCrawlerData<DataModels.TumblrSvcJson.Post>> jsonSvcQueue =
                 GetJsonQueue<DataModels.TumblrSvcJson.Post>();
             return new TumblrHiddenCrawler(shellService, crawlerService, webRequestFactory,
                 cookieService, GetTumblrDownloader(progress, blog, files, postQueue, pt, ct),
                 GetTumblrJsonDownloader(jsonSvcQueue, blog, pt, ct), GetTumblrSvcJsonToTextParser(blog), GetTumblrParser(),
                 imgurParser, gfycatParser, GetWebmshareParser(), GetMixtapeParser(), GetUguuParser(), GetSafeMoeParser(),
                 GetLoliSafeParser(), GetCatBoxParser(), postQueue, jsonSvcQueue, blog, progress, pt, ct);
         case BlogTypes.tlb:
             return new TumblrLikedByCrawler(shellService, crawlerService, webRequestFactory,
                 cookieService, GetTumblrDownloader(progress, blog, files, postQueue, pt, ct), GetTumblrParser(),
                 imgurParser, gfycatParser, GetWebmshareParser(), GetMixtapeParser(), GetUguuParser(),
                 GetSafeMoeParser(), GetLoliSafeParser(), GetCatBoxParser(), postQueue, blog, progress, pt, ct);
         case BlogTypes.tumblrsearch:
             IPostQueue<TumblrCrawlerData<DataModels.TumblrSearchJson.Datum>> jsonQueue = GetJsonQueue<DataModels.TumblrSearchJson.Datum>();
             return new TumblrSearchCrawler(shellService, crawlerService, webRequestFactory,
                 cookieService, GetTumblrDownloader(progress, blog, files, postQueue, pt, ct), GetTumblrJsonDownloader(jsonQueue, blog, pt, ct),
                 GetTumblrParser(), imgurParser, gfycatParser, GetWebmshareParser(), GetMixtapeParser(), GetUguuParser(),
                 GetSafeMoeParser(), GetLoliSafeParser(), GetCatBoxParser(), postQueue, jsonQueue, blog, progress, pt, ct);
         case BlogTypes.tumblrtagsearch:
             IPostQueue<TumblrCrawlerData<DataModels.TumblrTaggedSearchJson.Datum>> jsonTagSearchQueue =
                 GetJsonQueue<DataModels.TumblrTaggedSearchJson.Datum>();
             return new TumblrTagSearchCrawler(shellService, crawlerService, webRequestFactory,
                 cookieService, GetTumblrDownloader(progress, blog, files, postQueue, pt, ct),
                 GetTumblrJsonDownloader(jsonTagSearchQueue, blog, pt, ct), GetTumblrParser(),
                 imgurParser, gfycatParser, GetWebmshareParser(), GetMixtapeParser(), GetUguuParser(),
                 GetSafeMoeParser(), GetLoliSafeParser(), GetCatBoxParser(), postQueue, jsonTagSearchQueue, blog, progress, pt, ct);
         default:
             throw new ArgumentException("Website is not supported!", nameof(blog));
     }
 }
Exemple #30
0
 public TumblrHiddenCrawler(IShellService shellService, CancellationToken ct, PauseToken pt, IProgress <DownloadProgress> progress,
                            ICrawlerService crawlerService, IWebRequestFactory webRequestFactory, ISharedCookieService cookieService, IDownloader downloader,
                            ICrawlerDataDownloader crawlerDataDownloader, ITumblrToTextParser <Post> tumblrJsonParser, IImgurParser imgurParser,
                            IGfycatParser gfycatParser, IWebmshareParser webmshareParser, IPostQueue <TumblrPost> postQueue,
                            IPostQueue <TumblrCrawlerData <Post> > jsonQueue, IBlog blog)
     : base(shellService, ct, progress, webRequestFactory, cookieService, postQueue, blog)
 {
     this.crawlerService        = crawlerService;
     this.downloader            = downloader;
     this.pt                    = pt;
     this.tumblrJsonParser      = tumblrJsonParser;
     this.imgurParser           = imgurParser;
     this.gfycatParser          = gfycatParser;
     this.webmshareParser       = webmshareParser;
     this.jsonQueue             = jsonQueue;
     this.crawlerDataDownloader = crawlerDataDownloader;
 }
Exemple #31
0
        } // end method

        public async Task<byte[,,]> fdenoiseNeural3(byte[,,] noisyIm, int overlap, string fileName, int layer, int[] networkSize, int numberofsectors, CancellationToken cancelToken, PauseToken pauseToken)
        {
            /*
             *   noisyIm: an image corrupted by AWG noise
             *        the sliding window stride of the denoising 
             *        process (a smaller stride will usually provide better results).
             *        The pixels of the clean image are assumed to be approximately in 
             *        the range 0..255.
             *   overlap: the number of overlaps for two consecutive patches measured in pixels.
             *        Those pixels that are overlapped will be averaged.
             *   fileName: a string which specified the name of the weight file.
             *   layer: Number of layers of the network.
             *   networkSize: Defines the number of neurons in each layer
             *   inputsPerSample: defines the number of inputs in each neuron`
            */
            #region Initialization
            // determine number of samples and populate them
            int[] inputsPerSample = new int[layer];
            inputsPerSample[0] = networkSize[layer - 1] + 1;
            for (int i = 1; i < layer; i++)
                inputsPerSample[i] = networkSize[0] + 1;
            // end for
            window.SetText2("Using the Window Step method.\r\n" + Environment.NewLine);

            window.SetText2("Initializing Components..\r\n" + Environment.NewLine);
            window.SetProgress1(2);

            window.SetText2("Loading weights..." + Environment.NewLine);
            // load the weights
            Complex[][,] weights = loadMlmvnWeights(fileName, layer, networkSize, inputsPerSample);
            window.SetText2("Done." + Environment.NewLine);


            window.SetText2("Configuring Patch Size...");
            // size of input / output patch
            int patchSz = (int)Math.Sqrt(weights[0].GetLength(1));
            int patchSzOut = (int)Math.Sqrt(weights[layer - 1].GetLength(0));
            // Size of each sector on unit circle

            window.SetText2("Done.\r\n" + Environment.NewLine);
            window.SetText2("Input patch size is: " + patchSz + Environment.NewLine);
            window.SetText2("Output patch size is: " + patchSzOut + Environment.NewLine);

            // calculate the difference of the patches
            int p_diff = (patchSz - patchSzOut) / 2;
            // check if input is larger than output. If so, extend the image
            int height = noisyIm.GetLength(0);
            int origHeight = height;
            int width = noisyIm.GetLength(1);
            int origWidth = width;
            int samples = noisyIm.GetLength(2);
            if (p_diff > 0)
            {
                noisyIm = new byte[height + p_diff * 2, width + p_diff * 2, samples];
                noisyIm = functions.MirrorImage(noisyIm, height, width, samples, p_diff);
                // if extended the image, update the size
                height = noisyIm.GetLength(0);
                width = noisyIm.GetLength(1);
            }

            #region Patch range configuration
            // calculate how many pixels will be skipped for each patch
            int interval = patchSz - overlap;
            // offsetX and offsetY determine the number of "leftover" pixels on the right and bottom edges.
            int offsetX = (width - patchSz) % interval;
            int offsetY = (height - patchSz) % interval;
            // reserve the array to indicate the index of patches. include one position for the fist patch.  And reserve extra one position just in case 
            // we need to fill the offset
            int[] range_x = new int[(width - patchSz) / interval + 2];
            int[] range_y = new int[(height - patchSz) / interval + 2];
            int pos = 0;
            // fill the arrays with intervals.  ignore the last element because we don't know if it's necessary yet
            for (int i = 0; i < range_x.GetLength(0) - 1; i++)
            {
                range_x[i] = pos;
                pos += interval;
            }
            pos = 0;
            for (int i = 0; i < range_y.GetLength(0) - 1; i++)
            {
                range_y[i] = pos;
                pos += interval;
            }
            // end for

            // correct last index if necessary
            // if offsetX and Y are equal to 0, that means no fitting is necessary.  Therefore, just resize the array to have
            // one less length.  Else, fill the last element of the array with the index according to the offsets
            if (offsetX == 0)
                Array.Resize(ref range_x, range_x.GetLength(0) - 1);
            else
                range_x[range_x.GetLength(0) - 1] = width - patchSz;
            // end if
            if (offsetY == 0)
                Array.Resize(ref range_y, range_y.GetLength(0) - 1);
            else
                range_y[range_y.GetLength(0) - 1] = height - patchSz;
            // end if
            #endregion
            window.SetText2("\r\nDifference of the patche size is: " + p_diff + Environment.NewLine);
            window.SetText2("Beginning variable initialization...");

            // pre-instantiate complex 2d-arrays
            // patch of interest
            byte[,,] cleanIm = new byte[origHeight, origWidth, samples];
            byte[,,] counter = new byte[origHeight, origWidth, samples]; // counts the overlapped patch, then later store the processed image.
            double[,] inputArray = new double[patchSz, patchSz];
            Complex[,] CinputArray = new Complex[patchSz, patchSz];
            // output patch to be stored to actual image
            byte[,] outputArray = new byte[patchSz, patchSz];
            byte[] output = new byte[(int)Math.Pow(patchSz, 2)];
            // used when patch needs to be transformed to 1d array
            Complex[] S = new Complex[inputArray.Length];
            // store outputs of network
            Complex[][] outputNeurons = new Complex[layer][];
            double[] dOutputNeurons = new double[networkSize[layer - 1]];
            // instanciate a jagged array to store outputs
            for (int i = 0; i < layer; i++)
                outputNeurons[i] = new Complex[networkSize[i]];
            // end for
            Complex sum = new Complex(0, 0);
            S[0] = new Complex(1, 0);
            // instantiate imaginary unit
            Complex complex1 = new Complex(0.0, 1.0);
            // processIndex as in old code
            int offset = ((patchSzOut - 3) / 2) + 1;
            double bb = (2 * Math.PI) / numberofsectors;

            string[] Channels;

            switch (samples)
            {
                case 1:
                    Channels = new string[] { "" };
                    break;
                case 3:
                    Channels = new string[] { "red ", "green ", "blue " };
                    break;
                default:
                    Channels = new string[] { "" };
                    break;
            }

            window.SetText2("Done." + Environment.NewLine);
            window.SetText2("Beginning Processing.\r\n" + Environment.NewLine);
            #endregion
            // --------------- Processing Begins ------------------------------
            // process each samples
            for (int k = 0; k < samples; k++)
            {
                for (int row = 0; row < range_y.GetLength(0); row++) // for each row
                {
                    for (int col = 0; col < range_x.GetLength(0); col++) // for each column
                    {
                        #region process first layer
                        // process first layer
                        int ii = 0;
                        byte[,] src = Functions.CreatePatch(noisyIm, range_y[row], range_x[col], k, patchSz);
                        // upcast to double
                        Array.Copy(src, inputArray, src.Length);
                        // transformation of inputs into complex plane
                        for (int i = 0; i < patchSz; i++)
                            for (int j = 0; j < patchSz; j++)
                                CinputArray[i, j] = Exp(complex1 * 2 * Math.PI * inputArray[i, j] / numberofsectors);
                        // end nested for loop
                        // transform to 1d array
                        for (int i = 0; i < patchSz; i++)
                            for (int j = 0; j < patchSz; j++)
                                S[i * patchSz + j] = CinputArray[i, j];
                        // end for loop
                        #endregion
                        #region calculate weighted sum of first layer and its activation
                        // calculate weighted sum & activation
                        for (int i = 0; i < networkSize[0]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[0]; j++)
                            {
                                sum = sum + weights[ii][i, j] * S[j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            outputNeurons[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        // apply continuous activation
                        for (int t = 0; t < networkSize[ii]; t++)
                            outputNeurons[ii][t] /= Complex.Abs(outputNeurons[ii][t]);
                        // end for
                        #endregion
                        #region calculate weighted sum of second to last layer
                        // ----------------- Process second to last hidden layers, then output layer
                        for (ii = 1; ii < layer - 1; ii++)
                        {
                            for (int i = 0; i < networkSize[ii]; i++)
                            {
                                for (int j = 1; j < inputsPerSample[ii]; j++)
                                {
                                    sum = sum + weights[ii][i, j] * outputNeurons[ii - 1][j - 1];
                                }
                                sum = sum + weights[ii][i, 0];
                                outputNeurons[ii][i] = sum;
                                sum = new Complex(0, 0);
                            } // end for
                            // apply contiunous activation
                            for (int t = 0; t < networkSize[ii]; t++)
                                outputNeurons[ii][t] /= Complex.Abs(outputNeurons[ii][t]);
                            // end for
                        } // end for ii


                        // output layer
                        ii = layer - 1; // set to last layer
                        // calculate the weighted sum
                        for (int i = 0; i < networkSize[ii]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[ii]; j++)
                            {
                                sum = sum + weights[ii][i, j] * outputNeurons[ii - 1][j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            outputNeurons[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        for (int jj = 0; jj < networkSize[ii]; jj++)
                        {
                            // calculate discrete output
                            // get angle
                            dOutputNeurons[jj] = Math.Atan2(outputNeurons[ii][jj].Imaginary, outputNeurons[ii][jj].Real);
                            if (dOutputNeurons[jj] < 0)
                                dOutputNeurons[jj] = 2 * Math.PI + dOutputNeurons[jj];
                            // end if
                            // round
                            dOutputNeurons[jj] = Math.Truncate(dOutputNeurons[jj] / bb);
                            //dOutputNeurons[jj] = Math.Floor(dOutputNeurons[jj]/bb);

                            if (dOutputNeurons[jj] > 255)
                                if (dOutputNeurons[jj] < 320)
                                    dOutputNeurons[jj] = 255;
                                else
                                    dOutputNeurons[jj] = 0;
                            // end if 
                            // convert results to byte
                            output[jj] = Convert.ToByte(dOutputNeurons[jj]);
                        } // end for

                        #endregion second to last layer
                        #region Process image

                        //// resize
                        //for (int i = 0; i < patchSzOut; i++)
                        //    for (int j = 0; j < patchSzOut; j++)
                        //        outputArray[i, j] = output[p_diff + j + (i * patchSz)];
                        //// end for

                        // add to the actual image
                        for (int i = 0; i < patchSzOut; i++)
                            for (int j = 0; j < patchSzOut; j++)
                            {
                                //if (counter[range_y[row] + i, range_x[col] + j] == 0)
                                //{
                                cleanIm[range_y[row] + i, range_x[col] + j, k] += output[p_diff + j + (i * patchSz)];
                                // end for
                                //cleanIm[range_y[row] + i, range_x[col] + j, k] += outputArray[i, j];
                                counter[range_y[row] + i, range_x[col] + j, k]++;
                                //}
                            }
                        // end for

                        #endregion
                        #region Form elements
                        if (cancelToken.IsCancellationRequested)
                            cancelToken.ThrowIfCancellationRequested();

                        // Action when pause button is clicked
                        await pauseToken.WaitWhilePausedAsync();

                        window.SetProgress1(1);
                        #endregion
                    } // end col for loop
                    window.SetText2("Patches in " + Channels[k] + "row " + (row + 1) + " of " + range_y.Length + " done." + Environment.NewLine);
                } // end row for loop
            }
            #region Average
            // Average
            for (int row = 0; row < origHeight; row++) // for each row
            {
                for (int col = 0; col < origWidth; col++) // for each column
                {
                    for (int k = 0; k < samples; k++) // for each sample
                    {
                    cleanIm[row, col, k] /= counter[row, col, k];
                    counter[row, col, k] = Convert.ToByte(cleanIm[row, col, k]);
                    }
                }
            }
            #endregion
            return counter;
        }
Exemple #32
0
        public async Task<Complex[][,]> Learning(string fileNameSamples, int numberOfInputSamples, string fileNameWeights, int layer, int[] networkSize, int[] inputsPerSample, int numberofsectors, double globalThreasholdValue, double localThresholdValue, bool randomWeights, CancellationToken cancelToken, PauseToken pauseToken)
        {
            #region Initialization
            int twoInputsPerSample = networkSize[layer - 1] * 2;

            window.SetText1("Initializing components..." + Environment.NewLine);
            // load the samples
            window.SetText1("Loading learning samples... ");
            byte[,] samples = loadLearningSamples(fileNameSamples, numberOfInputSamples, twoInputsPerSample);
            window.SetText1("Done." + Environment.NewLine);
            // Initial Weights Initialization
            #region Weights Initialization
            Random random = new Random();
            double real;
            double imag;
            Complex[][,] weights = new Complex[layer][,];
            if (randomWeights)
            {
                // generate random weights
                // initialize weights matrix       
                for (int ii = 0; ii < layer; ii++)
                {
                    weights[ii] = new Complex[networkSize[ii], inputsPerSample[ii]];
                    //if (ii == 0)
                    //    weights[ii] = new Complex[networkSize[ii], inputsPerSample[ii]];
                    //else
                    //    weights[ii] = new Complex[networkSize[ii], networkSize[ii - 1]];
                    // now generate random numbers
                    for (int i = 0; i < weights[ii].GetLength(0); i++)
                        for (int j = 0; j < weights[ii].GetLength(1); j++)
                        {
                            real = random.NextDouble() - 0.5;
                            imag = random.NextDouble() - 0.5;
                            weights[ii][i, j] = new Complex(real, imag);
                        } // end for j
                    // end for i
                } // end for ii
            }
            else
            {
                window.SetText1("Loading weights... ");
                // load the weights
                weights = loadMlmvnWeights(fileNameWeights, layer, networkSize, inputsPerSample);
                window.SetText1("Done.\n" + Environment.NewLine);
            } // end if 
            #endregion
            double twoPi = Math.PI * 2;
            Complex complex1 = new Complex(0.0, 1.0);
            double sectorSize = twoPi / numberofsectors;
            Complex[] Sector = new Complex[numberofsectors];
            for (int i = 0; i < numberofsectors; i++)
            {
                double angSector = twoPi * i / numberofsectors;
                Sector[i] = Complex.Exp(complex1 * angSector);
            }
            int numberOfOutputs = networkSize[layer - 1];
            int rowInputs = samples.GetLength(0);
            int colInputs = samples.GetLength(1) / 2;
            // Desired outputs
            byte[,] desiredOutputs = new byte[rowInputs, colInputs];
            for (int i = 0; i < rowInputs; i++)
                for (int j = 0; j < colInputs; j++)
                    desiredOutputs[i, j] = samples[i, j + colInputs];
            // end for loops

            // Resized Inputs
            byte[,] inputs = new byte[rowInputs, colInputs];
            for (int i = 0; i < rowInputs; i++)
                for (int j = 0; j < colInputs; j++)
                    inputs[i, j] = samples[i, j];
            // end for loops
            Complex[,] Cinputs = new Complex[rowInputs, colInputs];
            int[,] networkOutputs = new int[rowInputs, colInputs];
            //for (int i = 0; i < rowInputs; i++)
            //    for (int j = 0; j < colInputs; j++)
            //        networkOutputs[i, j] = new Complex(0, 0);
            // end
            double[] networkErrors = new double[rowInputs];
            Complex[][] neuronOutputs = new Complex[layer][];
            // instanciate a jagged array to store outputs
            for (int i = 0; i < layer; i++)
                neuronOutputs[i] = new Complex[networkSize[i]];
            // end for
            double[] dNeuronOutputs = new double[networkSize[layer - 1]];
            Complex[][] neuronErrors = neuronOutputs;
            Complex[][] weightedSum = new Complex[layer][];
            // instanciate a jagged array to store outputs
            for (int i = 0; i < layer; i++)
                weightedSum[i] = new Complex[networkSize[i]];
            // end for
            Complex sum = new Complex(0, 0);
            // transformation of inputs into complex plane
            for (int i = 0; i < rowInputs; i++)
                for (int j = 0; j < colInputs; j++)
                    Cinputs[i, j] = Exp(complex1 * 2 * Math.PI * inputs[i, j] / numberofsectors);
            // end nested for loop
            // initialize error criteria
            double mse = 0;
            double rmse = 0;
            double[] learningRate;
            // check if learning is finished
            bool finishedLearning = false;
            // counts each cycle
            int iteration = 0;

            // output calculation
            Complex[,] temp;
            Complex[,] a1;
            Complex[] b1;
            Complex[] c1;
            Complex d1;
            Complex[,] e1;
            Complex[,] f1;

            window.SetText1("Beginning the learning of the weights...\r\n" + Environment.NewLine);
            #endregion

            #region RMSE ALGORITHM
            // repeats process until learning converges
            while (!finishedLearning)
            {
                // increment iteration
                iteration++;
                // --------------- BEGIN OUTPUT CALCULATION ------------------------------
                #region OUTPUT CALCULATION
                // process each samples
                for (int aa = 0; aa < numberOfInputSamples; aa++) // for each row
                {
                    #region calculate weighted sum of first layer and its activation
                    // process first layer
                    int ii = 0;

                    // calculate weighted sum & activation
                    for (int i = 0; i < networkSize[0]; i++)
                    {
                        for (int j = 1; j < inputsPerSample[0]; j++)
                        {
                            sum = sum + weights[ii][i, j] * Cinputs[aa, j - 1];
                        }
                        sum = sum + weights[ii][i, 0];
                        neuronOutputs[ii][i] = sum;
                        //weightedSum[ii][i] = sum;
                        sum = new Complex(0, 0);
                    } // end for

                    // apply continuous activation
                    for (int t = 0; t < networkSize[ii]; t++)
                        neuronOutputs[ii][t] /= Complex.Abs(neuronOutputs[ii][t]);
                    // end for
                    #endregion
                    #region calculate weighted sum of second to last layer
                    // ----------------- Process second to last hidden layers ---------------
                    for (ii = 1; ii < layer - 1; ii++)
                    {
                        for (int i = 0; i < networkSize[ii]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[ii]; j++)
                            {
                                sum = sum + weights[ii][i, j] * neuronOutputs[ii - 1][j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            neuronOutputs[ii][i] = sum;
                            //weightedSum[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for
                        // apply contiunous activation
                        for (int t = 0; t < networkSize[ii]; t++)
                            neuronOutputs[ii][t] /= Complex.Abs(neuronOutputs[ii][t]);
                        // end for
                    } // end for ii


                    // ----------------- Process output layer --------------------------------
                    ii = layer - 1; // set to last layer
                    // calculate the weighted sum
                    for (int i = 0; i < networkSize[ii]; i++)
                    {
                        for (int j = 1; j < inputsPerSample[ii]; j++)
                        {
                            sum = sum + weights[ii][i, j] * neuronOutputs[ii - 1][j - 1];
                        }
                        sum = sum + weights[ii][i, 0];
                        neuronOutputs[ii][i] = sum;
                        //weightedSum[ii][i] = sum;
                        sum = new Complex(0, 0);
                    } // end for

                    for (int jj = 0; jj < networkSize[ii]; jj++)
                    {
                        // calculate discrete output
                        // get angle
                        dNeuronOutputs[jj] = Math.Atan2(neuronOutputs[ii][jj].Imaginary, neuronOutputs[ii][jj].Real);
                        // if output is less than 0, add 2 pi to make it positive
                        if (dNeuronOutputs[jj] < 0)
                            dNeuronOutputs[jj] = 2 * Math.PI + dNeuronOutputs[jj];
                        // end if
                        // round
                        dNeuronOutputs[jj] = Math.Truncate(dNeuronOutputs[jj] / sectorSize);
                        //dOutputNeurons[jj] = Math.Floor(dOutputNeurons[jj]/bb);
                        // convert results to byte... did not work correctly, because it could be more than 255.  So let it be integer.
                        networkOutputs[aa, jj] = Convert.ToInt32(dNeuronOutputs[jj]);
                    } // end for
                    #endregion second to last layer
                } // end row for loop
                #endregion
                // -------------- END OUTPUT CALCULATION -----------------------------

                // -------------- BEGIN NET ERROR CALCULATION ------------------------
                #region GLOBAL ERROR CALCULATION
                // calculate NET error
                for (int aa = 0; aa < numberOfInputSamples; aa++)
                {
                    for (int i = 0; i < colInputs; i++)
                    {
                        networkErrors[aa] += Math.Pow((networkOutputs[aa, i] - desiredOutputs[aa, i]), 2);
                    }
                    networkErrors[aa] /= numberOfOutputs;
                    mse += networkErrors[aa];
                } // end for aa
                // calculate mse
                mse /= numberOfInputSamples;
                // calculate rmse
                rmse = Math.Sqrt(mse);
                window.SetText1("Iteration " + iteration + " done.          RMSE: " + rmse + Environment.NewLine);
                // Check if learning has converged
  //--->        window.TaskbarItemInfo.ProgressState = System.Windows.Shell.TaskbarItemProgressState.Normal;
                // Action when cancel button is clicked
                if (cancelToken.IsCancellationRequested)
                    cancelToken.ThrowIfCancellationRequested();

                // Action when pause button is clicked
                await pauseToken.WaitWhilePausedAsync();

                if (rmse <= globalThreasholdValue)
                {
                    finishedLearning = true;
                    window.SetText1("\r\nLearning Converged!!!" + Environment.NewLine);
                }
                // end if
                #endregion

                // --------------- BEGIN LEARNING / MODIFICATION OF WEIGHTS ---------------------------
                // LEARNING / MODIFICATION OF WEIGHTS
                // if the algorithm has not finished learning then output of the
                // network needs to be calculated again to start correction of
                // errors
                #region OUTPUT CALCULATION
                if (!finishedLearning)
                {

                    // calculating the output of the network for each sample and
                    // correcting weights if output is > localThresholdValue
                    for (int aa = 0; aa < numberOfInputSamples; aa++) // for each row
                    {
                        #region calculate weighted sum of first layer and its activation
                        // ii holds current layer index. Process first layer
                        int ii = 0;

                        // calculate weighted sum for 1st hidden layer
                        for (int i = 0; i < networkSize[ii]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[ii]; j++)
                            {
                                sum = sum + weights[ii][i, j] * Cinputs[aa, j - 1]; // <-- 7.2  of processing done here
                            }
                            sum = sum + weights[ii][i, 0];
                            //neuronOutputs[ii][i] = sum;
                            weightedSum[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        // apply continuous activation
                        for (int t = 0; t < networkSize[ii]; t++)
                            neuronOutputs[ii][t] = weightedSum[ii][t] / Complex.Abs(weightedSum[ii][t]);
                        // end for
                        #endregion
                        #region calculate weighted sum of second to last hidden layer
                        // ----------------- Process second to last hidden layers, then output layer
                        // ii holds current layer
                        for (ii = 1; ii < layer - 1; ii++)
                        {
                            for (int i = 0; i < networkSize[ii]; i++)
                            {
                                for (int j = 1; j < inputsPerSample[ii]; j++)
                                {
                                    sum = sum + weights[ii][i, j] * neuronOutputs[ii - 1][j - 1];
                                }
                                sum = sum + weights[ii][i, 0];
                                //neuronOutputs[ii][i] = sum;
                                weightedSum[ii][i] = sum;
                                sum = new Complex(0, 0);
                            } // end for
                            // apply continuous activation
                            for (int t = 0; t < networkSize[ii]; t++)
                                neuronOutputs[ii][t] = weightedSum[ii][t] / Complex.Abs(weightedSum[ii][t]);
                            // end for
                        } // end for ii
                        #endregion
                        #region calculate output layer and network output calculation
                        // output layer
                        ii = layer - 1; // set to last layer
                        // calculate the weighted sum
                        for (int i = 0; i < networkSize[ii]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[ii]; j++)
                            {
                                sum = sum + weights[ii][i, j] * neuronOutputs[ii - 1][j - 1]; // <-- 5.8 of processing is done here
                            }
                            sum = sum + weights[ii][i, 0];
                            neuronOutputs[ii][i] = sum;
                            weightedSum[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        // apply the activation function for discrete outputs
                        for (int jj = 0; jj < networkSize[ii]; jj++)
                        {
                            // Action when cancel button is clicked
                            if (cancelToken.IsCancellationRequested)
                                cancelToken.ThrowIfCancellationRequested();

                            // Action when pause button is clicked
                            await pauseToken.WaitWhilePausedAsync();

                            // calculate discrete output
                            // get angle
                            dNeuronOutputs[jj] = Math.Atan2(weightedSum[ii][jj].Imaginary, weightedSum[ii][jj].Real);
                            // if output is less than 0, add 2 pi to make it positive
                            if (dNeuronOutputs[jj] < 0)
                                dNeuronOutputs[jj] = 2 * Math.PI + dNeuronOutputs[jj];
                            // end if
                            // round
                            dNeuronOutputs[jj] = Math.Truncate(dNeuronOutputs[jj] / sectorSize);
                            //dOutputNeurons[jj] = Math.Floor(dOutputNeurons[jj]/bb);
                            // convert results to byte... did not work correctly, because it could be more than 255.  So let it be integer.
                            networkOutputs[aa, jj] = Convert.ToInt32(dNeuronOutputs[jj]);
                        } // end for
                        #endregion second to last layer

                        #region R/MSE SPECIFIC CALCULATION OF ERROR
                        // calculate NET error
                        for (int i = 0; i < colInputs; i++)
                        {
                            networkErrors[aa] += Math.Pow((networkOutputs[aa, i]) - desiredOutputs[aa, i], 2);
                        }
                        networkErrors[aa] /= numberOfOutputs;
                        // calculate rmse
                        networkErrors[aa] = Math.Sqrt(networkErrors[aa]);
                        #endregion

                        #region Weights Modification
                        // check agains local threshold
                        // if localThresholdValue is greater, then the weights are corrected
                        if (localThresholdValue < networkErrors[aa])
                        {
                            // Action when cancel button is clicked
                            if (cancelToken.IsCancellationRequested)
                                cancelToken.ThrowIfCancellationRequested();

                            // Action when pause button is clicked
                            await pauseToken.WaitWhilePausedAsync();

                            #region ERROR CALCULATION
                            // CALCULATE THE ERROR OF THE NEURONS
                            // calculation of the errors of neurons starts at last layer
                            // and moves to first layer
                            ii = layer - 1;
                            // outputs will contain normalized weighted sums for all output neurons
                            for (int t = 0; t < networkSize[ii]; t++)
                                neuronOutputs[ii][t] = weightedSum[ii][t] / Complex.Abs(weightedSum[ii][t]);
                            // end for
                            // the global error for the jjj-th output neuron
                            // equals a root of unity corresponding to the
                            // desired output - normalized weighted sum for
                            // the corresponding output neuron
                            for (int jjj = 0; jjj < networkSize[ii]; jjj++)
                                neuronErrors[ii][jjj] = Sector[desiredOutputs[aa, jjj]] - neuronOutputs[ii][jjj];
                            // end for
                            // finally we obtain the output neurons' errors
                            // normalizing the global errors (dividing them
                            // by the (number of neurons in the preceding
                            // layer+1)
                            for (int t = 0; t < networkSize[ii]; t++)
                                neuronErrors[ii][t] = neuronErrors[ii][t] / (networkSize[ii - 1] + 1);
                            // end for

                            // ------------- HANDLING THE REST OF LAYERS - ERROR BACKPROPAGATION ------------
                            for (ii = layer - 2; -1 < ii; ii--)
                            {
                                // calculate the reciprocal weights for the layer ii and putting them in 
                                // a vector-row 
                                temp = new Complex[weights[ii + 1].GetLength(1) - 1, weights[ii + 1].GetLength(0)];
                                for (int i = 1; i < temp.GetLength(0) + 1; i++) //511
                                    for (int j = 0; j < temp.GetLength(1); j++) //169
                                        temp[i - 1, j] = 1 / weights[ii + 1][j, i]; // <-- 23.5 of processing done here
                                //end fors
                                // confirmed bug free above this point... 5/2/2014 0:48

                                sum = new Complex(0, 0);
                                // backpropagation of weights
                                if (0 < ii) // all hidden layers except the 1st
                                    for (int row = 0; row < temp.GetLength(0); row++)
                                    {
                                        for (int col = 0; col < temp.GetLength(1); col++)
                                        {
                                            sum = sum + temp[row, col] * neuronErrors[ii + 1][col];
                                        }
                                        neuronErrors[ii][row] = sum / (networkSize[ii - 1] + 1);
                                        sum = new Complex(0, 0);
                                        // end for
                                    } // end for
                                else
                                    for (int row = 0; row < temp.GetLength(0); row++)
                                    {
                                        for (int col = 0; col < temp.GetLength(1); col++)
                                        {
                                            sum = sum + temp[row, col] * neuronErrors[ii + 1][col];
                                        }
                                        neuronErrors[ii][row] = sum / (inputsPerSample[0]);
                                        sum = new Complex(0, 0);
                                        // end for
                                    } // end for
                                // end if
                            } // end ii for loop over the layers
                            #endregion

                            #region WEIGHTS CORRECTION
                            // -------------- CORRECTS THE WEIGHTS OF THE NETWORK ----------------------------------
                            // HANDLING THE 1ST HIDDEN LAYER
                            // learning rate is a reciprocal absolute value of the weighted sum
                            learningRate = new double[weightedSum[0].GetLength(0)];
                            for (int i = 0; i < learningRate.GetLength(0); i++)
                                learningRate[i] = (1 / Complex.Abs(weightedSum[0][i]));
                            // end for

                            // take learning rate into account
                            // for all weights except bias w0
                            for (int row = 0; row < networkSize[0]; row++)
                                for (int col = 1; col < inputsPerSample[0]; col++)
                                    weights[0][row, col] = weights[0][row, col] + (learningRate[row] * neuronErrors[0][row]) * Complex.Conjugate(Cinputs[aa, col - 1]);
                            // end for
                            for (int row = 0; row < networkSize[0]; row++)
                                weights[0][row, 0] = weights[0][row, 0] + (learningRate[row] * neuronErrors[0][row]);
                            // end for


                            // correct the following layers
                            for (ii = 1; ii < layer; ii++)
                            {
                                sum = new Complex(0, 0);
                                // calculate the new output of preceding layer
                                // if preceding layer is the 1st one
                                if (ii == 1)
                                {
                                    // calculate weighted sum
                                    for (int i = 0; i < networkSize[0]; i++)
                                    {
                                        for (int j = 1; j < inputsPerSample[0]; j++)
                                        {
                                            sum = sum + weights[0][i, j] * Cinputs[aa, j - 1];
                                        }
                                        sum = sum + weights[0][i, 0];
                                        weightedSum[0][i] = sum;
                                        sum = new Complex(0, 0);
                                    } // end for
                                } // end if
                                else // if a preceding layer is not the 1st one
                                {
                                    // calculate weighted sum & activation
                                    for (int i = 0; i < networkSize[ii - 1]; i++)
                                    {
                                        for (int j = 1; j < inputsPerSample[ii - 1]; j++)
                                        {
                                            sum = sum + weights[ii - 1][i, j] * neuronOutputs[ii - 2][j - 1];
                                        }
                                        sum = sum + weights[ii - 1][i, 0];
                                        weightedSum[ii - 1][i] = sum;
                                        sum = new Complex(0, 0);
                                    } // end for
                                } // end if

                                // CONTINUOUS OUTPUT CALCULATION
                                for (int t = 0; t < networkSize[ii - 1]; t++)
                                    neuronOutputs[ii - 1][t] = weightedSum[ii - 1][t] / Complex.Abs(weightedSum[ii - 1][t]);
                                // end for

                                // learning rate is the reciprocal absolute value of the weighted sum
                                learningRate = new double[weightedSum[ii].GetLength(0)];
                                for (int i = 0; i < learningRate.GetLength(0); i++)
                                    learningRate[i] = (1 / Complex.Abs(weightedSum[ii][i]));
                                // end for

                                // learning rate not used for the output layer neurons
                                if (ii < layer) // <-- ~10  of processing done here (partly due to complex calculations)
                                {
                                    // do last things
                                    a1 = (Complex[,])weights[ii].Clone();
                                    b1 = (Complex[])neuronErrors[ii].Clone();
                                    for (int i = 0; i < b1.GetLength(0); i++)
                                        b1[i] = b1[i] * learningRate[i];
                                    // end for
                                    c1 = (Complex[])neuronOutputs[ii - 1].Clone();
                                    for (int i = 0; i < c1.GetLength(0); i++)
                                        c1[i] = Complex.Conjugate(c1[i]);
                                    // end for
                                    e1 = (Complex[,])a1.Clone();
                                    for (int i = 0; i < networkSize[ii]; i++)
                                    {
                                        d1 = b1[i];
                                        for (int j = 1; j < a1.GetLength(1); j++)
                                            e1[i, j] = d1 * c1[j - 1];
                                    }
                                    f1 = new Complex[a1.GetLength(0), a1.GetLength(1) - 1];
                                    for (int i = 0; i < a1.GetLength(0); i++)
                                        for (int j = 1; j < a1.GetLength(1); j++)
                                            f1[i, j - 1] = a1[i, j] + e1[i, j];
                                    // end for

                                    for (int i = 0; i < weights[ii].GetLength(0); i++)
                                        for (int j = 1; j < weights[ii].GetLength(1); j++)
                                            weights[ii][i, j] = f1[i, j - 1];
                                    // end for
                                    for (int i = 0; i < weights[ii].GetLength(0); i++)
                                        weights[ii][i, 0] = weights[ii][i, 0] + (learningRate[i] * neuronErrors[ii][i]);
                                    // end for
                                }
                                else
                                {
                                    // do last things
                                    a1 = (Complex[,])weights[ii].Clone();
                                    b1 = (Complex[])neuronErrors[ii].Clone();
                                    c1 = (Complex[])neuronOutputs[ii - 1].Clone();
                                    for (int i = 0; i < c1.GetLength(0); i++)
                                        c1[i] = Complex.Conjugate(c1[i]);
                                    // end for
                                    e1 = (Complex[,])a1.Clone();
                                    for (int i = 0; i < networkSize[ii]; i++)
                                    {
                                        d1 = b1[i];
                                        for (int j = 1; j < a1.GetLength(1); j++)
                                            e1[i, j] = d1 * c1[j - 1];
                                    }
                                    f1 = new Complex[a1.GetLength(0), a1.GetLength(1) - 1];
                                    for (int i = 0; i < a1.GetLength(0); i++)
                                        for (int j = 1; j < a1.GetLength(1); j++)
                                            f1[i, j - 1] = a1[i, j] + e1[i, j];
                                    // end for

                                    for (int i = 0; i < weights[ii].GetLength(0); i++)
                                        for (int j = 1; j < weights[ii].GetLength(1); j++)
                                            weights[ii][i, j] = f1[i, j - 1];
                                    // end for
                                    for (int i = 0; i < weights[ii].GetLength(0); i++)
                                        weights[ii][i, 0] = weights[ii][i, 0] + neuronErrors[ii][i];
                                    // end for
                                } // end if
                            } // end ii for
                            #endregion
                        #endregion
                        } // end localThresholdValue if check
                    } // end aa for loop
                } // end ~finishedLearning if statement
                #endregion
            }// end ~finishedLearning while loop
            #endregion

            return weights;
        }
Exemple #33
0
        /* -------------------------- Denoise by Pixels ----------------------------------------------------------------- */

        public async Task<byte[,,]> Activation(byte[,,] noisyImage, int kernel, string weights, int numberofsectors, int inLayerSize, int hidLayerSize, CancellationToken cancelToken, PauseToken pauseToken)
        {
            // get height and width
            int height = noisyImage.GetLength(0);
            int width = noisyImage.GetLength(1);
            int samples = noisyImage.GetLength(2);
            int offset;
            switch (kernel)
            {
                case 3:
                    offset = 1;
                    break;
                case 5:
                    offset = 2;
                    break;
                case 7:
                    offset = 3;
                    break;
                default:
                    offset = 0;
                    break;
            } // end switch
            if (offset == 0)
            {
                window.SetText1("Value of kernel is not properly set." + Environment.NewLine);
                return null;
            }

            // extend the image
            byte[,,] image = new byte[height + offset * 2, width + offset * 2, samples];
            image = functions.MirrorImage(noisyImage, height, width, samples, offset);

            window.SetProgress1(4);

            // pre-instantiate complex 2d-array
            double[,] inputArray = new double[kernel, kernel];
            var CinputArray = new Complex[kernel, kernel];
            // instantiate imaginary unit
            var complex1 = new Complex(0.0, 1.0);
            // pass to neural network

            byte[,] src;

            string[] Channels;

            switch (samples)
            {
                case 1:
                    Channels = new string[] { "" };
                    break;
                case 3:
                    Channels = new string[] { "red ", "green ", "blue " };
                    break;
                default:
                    Channels = new string[] { "" };
                    break;
            }

            //int FirstQ = offset;
            //Parallel.For(FirstQ, height + offset, Q =>
            for (int k = 0; k < samples; k++)
            {
                for (int Q = offset; Q < height + offset; Q++)
                {

                    for (int P = offset; P < width + offset; P++)
                    {

                        src = Functions.CreateWindow(image, Q, P, k, kernel, offset);
                        Array.Copy(src, inputArray, src.Length);
                        // transformation of inputs into complex plane
                        for (int i = 0; i < kernel; i++)
                        {
                            for (int j = 0; j < kernel; j++)
                                CinputArray[i, j] = Exp(complex1 * 2 * Math.PI * inputArray[i, j] / numberofsectors);
                        } // end nested for loop
                        // process
                        noisyImage[Q - offset, P - offset, k] = NeuralNetwork(CinputArray, weights, numberofsectors, inLayerSize, hidLayerSize);


                    }

                    // Action when cancel button is clicked
                    if (cancelToken.IsCancellationRequested)
                        cancelToken.ThrowIfCancellationRequested();

                    // Action when pause button is clicked
                    await pauseToken.WaitWhilePausedAsync();

                    //Increments progress bar
                    window.SetProgress1(1);

                    // Writes progress to console
                    if (Q % 10 == 0) window.SetText2("Pixels in " + Channels[k] + "row " + (Q) + " of " + height + " done." + Environment.NewLine);
                    if (Q == height && Q % 10 != 0) window.SetText2("Pixels in " + Channels[k] + "row " + (Q) + " of " + height + " done." + Environment.NewLine);
                }
            };//);  
            return noisyImage;
        }// end method
Exemple #34
0
        } // end method

        public async Task<byte[,,]> fdenoiseNeural2(byte[,,] noisyIm, int step, string fileName, int layer, int[] networkSize, int[] inputsPerSample, int numberofsectors, CancellationToken cancelToken, PauseToken pauseToken)//, double progressBar1, double progressBar1Max)
        {
            /*
            *   noisyIm: an image corrupted by AWG noise
                  * the sliding window stride of the denoising 
                    process (a smaller stride will usually provide better results).
                The pixels of the clean image are assumed to be approximately in 
                the range 0..255.
            */

            #region Initialization

            window.SetText2("Using the new patch method.\r\n" + Environment.NewLine);

            window.SetText2("Initializing Components...\r\n" + Environment.NewLine);
            window.SetProgress1(2);

            window.SetText2("Loading weights... ");
            // load the weights
            Complex[][,] weights = loadMlmvnWeights(fileName, layer, networkSize, inputsPerSample);
            window.SetText2("Done." + Environment.NewLine);

            window.SetText2("Configuring Patch Size... ");
            // size of input / output patch
            int patchSz = (int)Math.Sqrt(weights[0].GetLength(1));
            int patchSzOut = (int)Math.Sqrt(weights[layer - 1].GetLength(0));
            // Size of each sector on unit circle

            window.SetText2("Done.\r\n" + Environment.NewLine);

            window.SetText2("Input patch size is: " + patchSz + Environment.NewLine);
            window.SetText2("Output patch size is: " + patchSzOut + Environment.NewLine);

            // calculate the difference of the patches
            int p_diff = (patchSz - patchSzOut) / 2;
            // check if input is larger than output. If so, extend the image
            int height = noisyIm.GetLength(0);
            int origHeight = height;
            int width = noisyIm.GetLength(1);
            int origWidth = width;
            int samples = noisyIm.GetLength(2);
            if (p_diff > 0)
            {
                noisyIm = new byte[height + p_diff * 2, width + p_diff * 2, samples];
                noisyIm = functions.MirrorImage(noisyIm, height, width, samples, p_diff);
                // if extended the image, update the size
                height = noisyIm.GetLength(0);
                width = noisyIm.GetLength(1);
            }

            #region Patch range configuration

            // interval determines how many pixels would be skipped before new patch will be placed.
            // For example, if step is 3 and patch size is 13, 3 * 2 = 6 pixels would be overlapped for non-edge patches.
            // Therefore, 13 - 6 = 7 pixels will be skipped.
            int interval = patchSz - (step * 2);

            // offsetX and offsetY determine the number of "leftover" pixels on the right and bottom edges.
            // For example, if 512 * 512 image will be filled by 13 * 13 patches, 512 - ((13-3) % interval
            int offsetX = (width - (patchSz - step)) % interval;
            int offsetY = (height - (patchSz - step)) % interval;
            // reserve the array to indicate the index of patches. include one position for the fist patch.  And reserve extra one position just in case 
            // we need to fill the offset
            int[] range_x = new int[(width - (patchSz - step)) / interval + 2];
            int[] range_y = new int[(height - (patchSz - step)) / interval + 2];
            int pos = 0;
            // fill the arrays with intervals.  ignore the last element because we don't know if it's necessary yet
            for (int i = 0; i < range_x.GetLength(0) - 1; i++)
            {
                range_x[i] = pos;
                pos += interval;
            }
            pos = 0;
            for (int i = 0; i < range_y.GetLength(0) - 1; i++)
            {
                range_y[i] = pos;
                pos += interval;
            }
            // end for

            // correct last index if necessary
            // if offsetX and Y are equal to 0, that means no fitting is necessary.  Therefore, just resize the array to have
            // one less length.  Else, fill the last element of the array with the index according to the offsets
            if (offsetX == 0)
                Array.Resize(ref range_x, range_x.GetLength(0) - 1);
            else
                range_x[range_x.GetLength(0) - 1] = width - patchSz;
            // end if
            if (offsetY == 0)
                Array.Resize(ref range_y, range_y.GetLength(0) - 1);
            else
                range_y[range_y.GetLength(0) - 1] = height - patchSz;
            // end if
            #endregion

            window.SetText2("\r\nDifference of the patche size is: " + p_diff + Environment.NewLine);
            window.SetText2("Beginning variable initialization... ");

            // pre-instantiate complex 2d-arrays
            // patch of interest
            byte[,,] cleanIm = new byte[origHeight, origWidth, samples];
            //byte[,] counter = new byte[origHeight, origWidth]; // counts the overlapped patch, then later store the processed image.
            double[,] inputArray = new double[patchSz, patchSz];
            Complex[,] CinputArray = new Complex[patchSz, patchSz];
            // output patch to be stored to actual image
            byte[,] outputArray = new byte[patchSz, patchSz];
            byte[] output = new byte[Convert.ToInt32(Math.Pow(patchSz, 2))];
            // used when patch needs to be transformed to 1d array
            Complex[] S = new Complex[inputArray.Length];
            // store outputs of network
            Complex[][] outputNeurons = new Complex[layer][];
            double[] dOutputNeurons = new double[networkSize[layer - 1]];
            // instanciate a jagged array to store outputs
            for (int i = 0; i < layer; i++)
                outputNeurons[i] = new Complex[networkSize[i]];
            // end for
            Complex sum = new Complex(0, 0);
            S[0] = new Complex(1, 0);
            // instantiate imaginary unit
            Complex complex1 = new Complex(0.0, 1.0);
            // processIndex as in old code
            int offset = ((patchSzOut - 3) / 2) + 1;
            double bb = (2 * Math.PI) / numberofsectors;

            string[] Channels;

            switch (samples)
            {
                case 1:
                    Channels = new string[] {""};
                    break;
                case 3:
                    Channels = new string[] {"red ", "green ", "blue "};
                    break;
                default:
                    Channels = new string[] { "" };
                    break;
            }

            window.SetText2("Done.\r\n" + Environment.NewLine);
            window.SetText2("Beginning Processing... \r\n" + Environment.NewLine);
            #endregion

            // --------------- Processing Begins  ------------------------------
            // process each samples
            for (int k = 0; k < samples; k++)
            {
                for (int row = 0; row < range_y.GetLength(0); row++) // for each row
                {
                    for (int col = 0; col < range_x.GetLength(0); col++) // for each column
                    {
                        #region process first layer
                        // process first layer
                        int ii = 0;
                        byte[,] src = Functions.CreatePatch(noisyIm, range_y[row], range_x[col], k, patchSz);
                        // upcast to double
                        Array.Copy(src, inputArray, src.Length);
                        // transformation of inputs into complex plane
                        for (int i = 0; i < patchSz; i++)
                            for (int j = 0; j < patchSz; j++)
                                CinputArray[i, j] = Exp(complex1 * 2 * Math.PI * inputArray[i, j] / numberofsectors);
                        // end nested for loop
                        // transform to 1d array
                        for (int i = 0; i < patchSz; i++)
                            for (int j = 0; j < patchSz; j++)
                                S[i * patchSz + j] = CinputArray[i, j];
                        // end for loop
                        #endregion
                        #region calculate weighted sum of first layer and its activation
                        // calculate weighted sum & activation
                        for (int i = 0; i < networkSize[0]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[0]; j++)
                            {
                                sum = sum + weights[ii][i, j] * S[j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            outputNeurons[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        // apply continuous activation
                        for (int t = 0; t < networkSize[ii]; t++)
                            outputNeurons[ii][t] /= Complex.Abs(outputNeurons[ii][t]);
                        // end for
                        #endregion
                        #region calculate weighted sum of second to last layer
                        // ----------------- Process second to last hidden layers, then output layer
                        for (ii = 1; ii < layer - 1; ii++)
                        {
                            for (int i = 0; i < networkSize[ii]; i++)
                            {
                                for (int j = 1; j < inputsPerSample[ii]; j++)
                                {
                                    sum = sum + weights[ii][i, j] * outputNeurons[ii - 1][j - 1];
                                }
                                sum = sum + weights[ii][i, 0];
                                outputNeurons[ii][i] = sum;
                                sum = new Complex(0, 0);
                            } // end for
                            // apply contiunous activation
                            for (int t = 0; t < networkSize[ii]; t++)
                                outputNeurons[ii][t] /= Complex.Abs(outputNeurons[ii][t]);
                            // end for
                        } // end for ii


                        // output layer
                        ii = layer - 1; // set to last layer
                        // calculate the weighted sum
                        for (int i = 0; i < networkSize[ii]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[ii]; j++)
                            {
                                sum = sum + weights[ii][i, j] * outputNeurons[ii - 1][j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            outputNeurons[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        for (int jj = 0; jj < networkSize[ii]; jj++)
                        {
                            // calculate discrete output
                            // get angle
                            dOutputNeurons[jj] = Math.Atan2(outputNeurons[ii][jj].Imaginary, outputNeurons[ii][jj].Real);
                            if (dOutputNeurons[jj] < 0)
                                dOutputNeurons[jj] = 2 * Math.PI + dOutputNeurons[jj];
                            // end if
                            // round
                            dOutputNeurons[jj] = Math.Truncate(dOutputNeurons[jj] / bb);
                            //dOutputNeurons[jj] = Math.Floor(dOutputNeurons[jj]/bb);

                            if (dOutputNeurons[jj] > 255)
                                if (dOutputNeurons[jj] < 320)
                                    dOutputNeurons[jj] = 255;
                                else
                                    dOutputNeurons[jj] = 0;
                            // end if 
                            // convert results to byte
                            output[jj] = Convert.ToByte(dOutputNeurons[jj]);
                        } // end for

                        #endregion second to last layer
                        #region Process image
                        // resize
                        for (int i = 0; i < patchSzOut; i++)
                            for (int j = 0; j < patchSzOut; j++)
                                outputArray[i, j] = output[p_diff + j + (i * patchSz)];
                        // end for

                        // Output to an acutal image.
                        /* Codes below outputs the calculated patch into the output image.  Although the size of patch is 13 x 13, only center 7 * 7 will be copied to the
                         * image because of overlapping method, with the only exception of when the patch touches the edge of the image.  In order to determine this, we will
                         * first check whether the coordinate of the patch is either 0 or width(height) - patchSz. if so, we need to take the borders into account.  Otherwise,
                         * we just need to copy center 7 x 7 pixels onto corresponding coordinates.
                         */

                        // check
                        if (range_y[row] == 0 || range_y[row] == height - patchSz || range_x[col] == 0 || range_x[col] == width - patchSz)
                        {
                            // startY and startX determines the starting coordinate of the local patch; it's initialized with step. if the patch is touching the edge, we need to
                            // set them to 0, so whole patch side will be copied. The same for endY and endX.
                            int startY, startX;
                            startY = startX = step;
                            int endY, endX;
                            endY = endX = patchSzOut - step;
                            // All outer patches
                            if (range_y[row] == 0)
                                startY = 0;
                            // end if
                            if (range_y[row] == height - patchSz)
                            {
                                // startY = patchSzOut - (height - (range_y[row - 1] + patchSz - (step * 2)));
                                startY = patchSzOut - offsetY;
                                endY = patchSzOut;
                            }
                            // end if

                            if (range_x[col] == 0)
                                startX = 0;
                            // end if
                            if (range_x[col] == width - patchSz)
                            {
                                startX = patchSzOut - offsetX;
                                endX = patchSzOut;
                            }
                            // end if

                            // Place patches
                            for (int i = startY; i < endY; i++)
                            {
                                for (int j = startX; j < endX; j++)
                                {
                                    cleanIm[range_y[row] + i, range_x[col] + j, k] = outputArray[i, j];
                                } // end for
                            } //end
                        }
                        else
                        {
                            // All Inner Patches have outer edges cut off
                            for (int i = step; i < patchSzOut - step; i++)
                            {
                                for (int j = step; j < patchSzOut - step; j++)
                                {
                                    cleanIm[range_y[row] + i, range_x[col] + j, k] = outputArray[i, j];
                                } // end for
                            } //end
                        }

                        #endregion
                        #region Form elements
                        if (cancelToken.IsCancellationRequested)
                            cancelToken.ThrowIfCancellationRequested();

                        // Action when pause button is clicked
                        await pauseToken.WaitWhilePausedAsync();

                        window.SetProgress1(1);
                        #endregion
                    } // end col for loop               

                    window.SetText2("Patches in " + Channels[k] + "row " + (row + 1) + " of " + range_y.Length + " done." + Environment.NewLine);
                } // end row for loop
            }

            return cleanIm;
        } // end method
Exemple #35
0
        /* -------------------------- Denoise by Patch ----------------------------------------------------------------- */

        public async Task<byte[,,]> fdenoiseNeural(byte[,,] noisyIm, int step, string fileName, int layer, int[] networkSize, int[] inputsPerSample, int numberofsectors, CancellationToken cancelToken, PauseToken pauseToken)
        {
            /*
                noisyIm: an image corrupted by AWG noise
                    the sliding window stride of the denoising 
                    process (a smaller stride will usually provide better results).
                The pixels of the clean image are assumed to be approximately in 
                the range 0..255.
            */

            #region Initialization
            window.SetText2("Initializing components...\r\n" + Environment.NewLine); // Appends the text to the console window
            window.SetProgress1(2); // Increments the progress bar by 2

            window.SetText2("Loading weights... ");
            // load the weights
            Complex[][,] weights = loadMlmvnWeights(fileName, layer, networkSize, inputsPerSample);
            window.SetText2("Done." + Environment.NewLine);

            window.SetText2("Configuring patch size... ");

            // size of input / output patch
            int patchSz = (int)Math.Sqrt(weights[0].GetLength(1) - 1); // <-- Implement outside of function to determine type of weights
            int patchSzOut = (int)Math.Sqrt(weights[layer - 1].GetLength(0));
            // Size of each sector on unit circle
            window.SetText2("Done.\r\n" + Environment.NewLine);

            window.SetText2("Input patch size is: " + patchSz + Environment.NewLine);
            window.SetText2("Output patch size is: " + patchSzOut + Environment.NewLine);

            // calculate the difference of the patches
            int p_diff = (patchSz - patchSzOut) / 2;
            // check if input is larger than output. If so, extend the image
            int height = noisyIm.GetLength(0);
            int origHeight = height;
            int width = noisyIm.GetLength(1);
            int origWidth = width;
            int samples = noisyIm.GetLength(2);
            if (p_diff > 0)
            {
                noisyIm = new byte[height + p_diff * 2, width + p_diff * 2, samples];
                noisyIm = functions.MirrorImage(noisyIm, height, width, samples, p_diff);
                // if extended the image, update the size
                height = noisyIm.GetLength(0);
                width = noisyIm.GetLength(1);
            }

            #region Patch range configuration
            int pos = 0;
            // create arrays that contain the index ranges for row and column
            int[] range_y = new int[(height - patchSz) / step + 2];
            int[] range_x = new int[(width - patchSz) / step + 2];
            for (int i = 0; i < height - patchSz; i = i + step)
            {
                range_y[pos] = i;
                pos++;
            }
            // end for
            pos = 0;
            for (int i = 0; i < width - patchSz; i = i + step)
            {
                range_x[pos] = i;
                pos++;
            }
            // end for
            if (range_y[range_y.Length - 2] != height - patchSz)
            {
                range_y[range_y.Length - 1] = height - patchSz;
            }
            else
                Array.Resize(ref range_y, range_y.GetLength(0) - 1);
            // end if
            if (range_x[range_x.Length - 2] != height - patchSz)
            {
                range_x[range_x.Length - 1] = height - patchSz;
            }
            else
                Array.Resize(ref range_x, range_x.GetLength(0) - 1);
            // end if
            #endregion

            window.SetText2("\r\nDifference of the patche size is: " + p_diff + Environment.NewLine);
            window.SetText2("Beginning variable initialization... ");

            // pre-instantiate complex 2d-arrays
            // patch of interest
            int[,,] cleanIm = new int[origHeight, origWidth, samples];
            byte[,,] counter = new byte[origHeight, origWidth, samples]; // counts the overlapped patch, then later store the processed image.
            double[,] inputArray = new double[patchSz, patchSz];
            Complex[,] CinputArray = new Complex[patchSz, patchSz];
            // output patch to be stored to actual image
            byte[,] outputArray = new byte[patchSz, patchSz];
            byte[] output = new byte[(int)Math.Pow(patchSz, 2)];
            // used when patch needs to be transformed to 1d array
            Complex[] S = new Complex[inputArray.Length];
            // store outputs of network
            Complex[][] outputNeurons = new Complex[layer][];
            double[] dOutputNeurons = new double[networkSize[layer - 1]];
            // instanciate a jagged array to store outputs
            for (int i = 0; i < layer; i++)
                outputNeurons[i] = new Complex[networkSize[i]];
            // end for
            Complex sum = new Complex(0, 0);
            S[0] = new Complex(1, 0);
            // instantiate imaginary unit
            Complex complex1 = new Complex(0.0, 1.0);
            // processIndex as in old code
            int offset = ((patchSzOut - 3) / 2) + 1;
            double bb = (2 * Math.PI) / numberofsectors;

            string[] Channels;

            switch (samples)
            {
                case 1:
                    Channels = new string[] { "" };
                    break;
                case 3:
                    Channels = new string[] { "red ", "green ", "blue " };
                    break;
                default:
                    Channels = new string[] { "" };
                    break;
            }

            window.SetText2("Done.\r\n" + Environment.NewLine);
            window.SetText2("Beginning the processing... \r\n" + Environment.NewLine);
            #endregion

            //int increment = 0;
            //double test2 = range_x.GetLength(0)/2;
            //double test3 = Math.Floor(test2);
            // --------------- Processing Begins ------------------------------
            // process each samples

            for (int k = 0; k < samples; k++)
            {
                for (int row = 0; row < range_y.GetLength(0); row++) // for each row
                {
                    for (int col = 0; col < range_x.GetLength(0); col++) // for each column
                    {

                        #region process first layer
                        // process first layer
                        int ii = 0;
                        byte[,] src = Functions.CreatePatch(noisyIm, range_y[row], range_x[col], k, patchSz);
                        // upcast to double
                        Array.Copy(src, inputArray, src.Length);
                        // transformation of inputs into complex plane Inneficient?
                        for (int i = 0; i < patchSz; i++)
                            for (int j = 0; j < patchSz; j++)
                                CinputArray[i, j] = Exp(complex1 * 2 * Math.PI * inputArray[i, j] / numberofsectors);
                        // end nested for loop
                        // transform to 1d array
                        for (int i = 0; i < patchSz; i++)
                            for (int j = 0; j < patchSz; j++)
                                S[i * patchSz + j] = CinputArray[i, j];
                        // end for loop
                        #endregion
                        #region calculate weighted sum of first layer and its activation, line on MATLAB 1283
                        // calculate weighted sum & activation
                        for (int i = 0; i < networkSize[0]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[0]; j++)
                            {
                                sum = sum + weights[ii][i, j] * S[j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            outputNeurons[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        // apply continuous activation
                        for (int t = 0; t < networkSize[ii]; t++)
                            outputNeurons[ii][t] /= Complex.Abs(outputNeurons[ii][t]);
                        // end for
                        #endregion
                        #region calculate weighted sum of second to last layer
                        // ----------------- Process second to last hidden layers, then output layer
                        for (ii = 1; ii < layer - 1; ii++)
                        {
                            for (int i = 0; i < networkSize[ii]; i++)
                            {
                                for (int j = 1; j < inputsPerSample[ii]; j++)
                                {
                                    sum = sum + weights[ii][i, j] * outputNeurons[ii - 1][j - 1];
                                }
                                sum = sum + weights[ii][i, 0];
                                outputNeurons[ii][i] = sum;
                                sum = new Complex(0, 0);
                            } // end for
                            // apply contiunous activation
                            for (int t = 0; t < networkSize[ii]; t++)
                                outputNeurons[ii][t] /= Complex.Abs(outputNeurons[ii][t]);
                            // end for
                        } // end for ii


                        // output layer
                        ii = layer - 1; // set to last layer
                        // calculate the weighted sum
                        for (int i = 0; i < networkSize[ii]; i++)
                        {
                            for (int j = 1; j < inputsPerSample[ii]; j++)
                            {
                                sum = sum + weights[ii][i, j] * outputNeurons[ii - 1][j - 1];
                            }
                            sum = sum + weights[ii][i, 0];
                            outputNeurons[ii][i] = sum;
                            sum = new Complex(0, 0);
                        } // end for

                        for (int jj = 0; jj < networkSize[ii]; jj++)
                        {
                            // calculate discrete output
                            // get angle
                            dOutputNeurons[jj] = Math.Atan2(outputNeurons[ii][jj].Imaginary, outputNeurons[ii][jj].Real);
                            if (dOutputNeurons[jj] < 0)
                                dOutputNeurons[jj] = 2 * Math.PI + dOutputNeurons[jj];
                            // end if
                            // round
                            dOutputNeurons[jj] = Math.Truncate(dOutputNeurons[jj] / bb);
                            //dOutputNeurons[jj] = Math.Floor(dOutputNeurons[jj]/bb);

                            if (dOutputNeurons[jj] > 255)
                                if (dOutputNeurons[jj] < 320)
                                    dOutputNeurons[jj] = 255;
                                else
                                    dOutputNeurons[jj] = 0;
                            // end if 
                            // convert results to byte
                            output[jj] = Convert.ToByte(dOutputNeurons[jj]); //Trouble?
                        } // end for

                        #endregion second to last layer
                        #region Process image
                        // resize
                        for (int i = 0; i < patchSzOut; i++)
                            for (int j = 0; j < patchSzOut; j++)
                                outputArray[i, j] = output[p_diff + j + (i * patchSz)];
                        // end for
                        // add to the actual image
                        for (int i = 0; i < patchSzOut; i++)
                            for (int j = 0; j < patchSzOut; j++)
                            {
                                //if (counter[range_y[row] + i, range_x[col] + j] == 0)
                                //{
                                cleanIm[range_y[row] + i, range_x[col] + j, k] += outputArray[i, j];
                                counter[range_y[row] + i, range_x[col] + j, k]++;
                                //}
                            }
                        // end for
                        // end for
                        #endregion
                        #region Form elements
                        if (cancelToken.IsCancellationRequested)
                            cancelToken.ThrowIfCancellationRequested();

                        // Action when pause button is clicked
                        await pauseToken.WaitWhilePausedAsync();

                        window.SetProgress1(1);
                        #endregion

                    }//); // end col for loop

                    window.SetText2("Patches in " + Channels[k] + "row " + (row + 1) + " of " + range_y.Length + " done." + Environment.NewLine);

                }//); // end row for loop
            }
            #region Average
            // Average
            for (int row = 0; row < origHeight; row++) // for each row
            {
                for (int col = 0; col < origWidth; col++) // for each column
                {
                    for (int k = 0; k < samples; k++) // for each sample
                    {
                        cleanIm[row, col, k] /= counter[row, col, k];
                        counter[row, col, k] = Convert.ToByte(cleanIm[row, col, k]);
                    }
                }
            }
            #endregion

            #region Old code
            //double[,] pixel_weights = new double[patchSzOut, patchSzOut];
            //// median pixel ceiling(17/2) = 9
            //int mid = (int)Math.Ceiling((double)patchSzOut/2);
            //// floor(17/2) / 2 = 4
            //int sig = (int)Math.Floor((double)patchSzOut / 2) / weightSig;
            //// initialize pixel_weights
            //double d = 0;
            //for (int i = 0; i < patchSzOut; i++)
            //    for (int j = 0; j < patchSzOut; j++)
            //    {
            //        d = Math.Sqrt(Math.Pow((i - mid),2) + Math.Pow((j - mid), 2));
            //        pixel_weights[i, j] = Math.Exp(Math.Pow(-d,2) / (2 * Math.Pow(sig, 2))) / (sig * Math.Sqrt(2 * Math.PI));
            //    }
            //// end for
            //// obtain the max of pixel_weights and subtract each element with it
            //// to achieve mean zero
            //double max = pixel_weights.Cast<double>().Max();
            //for (int i = 0; i < patchSzOut; i++)
            //    for (int j = 0; j < patchSzOut; j++)
            //        pixel_weights[i, j] = pixel_weights[i, j] - max;
            //// end for

            //// Upcast byte to double
            //double[,] dnoisyIm = new double[noisyIm.GetLength(0), noisyIm.GetLength(1)];
            //Array.Copy(noisyIm, dnoisyIm, noisyIm.Length);
            //// subract 0.5 and multiply by 0.2 to achieve approximately mean zero and 
            //// variance close to one
            //for (int i = 0; i < dnoisyIm.GetLength(0); i++)
            //    for (int j = 0; j < dnoisyIm.GetLength(1); j++ )
            //        dnoisyIm[i, j] = (((dnoisyIm[i, j] / 255) - 0.5) / 0.2);
            //// end for

            //int chunkSize = 1000;
            //int pos = 0;
            //// create arrays that contain the index ranges for row and column
            //int[] range_y = new int[(dnoisyIm.GetLength(0) - patchSz)/step + 2];
            //int[] range_x = new int[(dnoisyIm.GetLength(1) - patchSz)/step + 2];
            //for (int i = 0; i < noisyIm.GetLength(0) - patchSz; i = i + step)
            //{
            //    range_y[pos] = i;
            //    pos++;
            //}
            //// end for
            //pos = 0;
            //for (int i = 0; i < noisyIm.GetLength(1) - patchSz; i = i + step)
            //{ 
            //    range_x[pos] = i;
            //    pos++;
            //}
            //// end for

            //// if the ranges do not include the last available row / column, include them
            //if (range_y[range_y.GetLength(0) - 2] != noisyIm.GetLength(0) - patchSz)
            //    range_y[range_y.GetLength(0) - 1] = noisyIm.GetLength(0) - patchSz;
            //else
            //    Array.Resize(ref range_y, range_y.GetLength(0) - 1);
            //// end if
            //if (range_x[range_x.GetLength(0) - 2] != noisyIm.GetLength(1) - patchSz)
            //    range_x[range_y.GetLength(0) - 1] = noisyIm.GetLength(1) - patchSz;
            //else
            //    Array.Resize(ref range_x, range_x.GetLength(0) - 1);
            //// end if

            //double[,] res = new double[(int)Math.Pow(patchSz, 2), chunkSize];
            //double[,] part = new double[(int)Math.Pow(patchSz, 2) + layer - 1, chunkSize];
            //int[,] positions_out = new int[2, chunkSize];
            //byte[,] denoisedIm = new byte[noisyIm.GetLength(0),noisyIm.GetLength(1)];
            //Complex[,] wIm = new Complex[noisyIm.GetLength(0), noisyIm.GetLength(1)];
            //byte[,] p = new byte[patchSz, patchSz];
            //byte[] p1 = new byte[(int)Math.Pow(patchSz,2)];

            // --------------------- Processing Image Begin ---------------------------------------
            //int idx = -1;
            //for (int y = 0; y < range_y.GetLength(0); y++)
            //{
            //    for (int x = 0; x < range_x.GetLength(0); x++)
            //    {
            //        // copy particular input patch from noisy image
            //        for (int i = 0; i < patchSz; i++)
            //            for (int j = 0; j < patchSz; j++)
            //                p[i, j] = noisyIm[range_y[y] + i, range_x[x] + j];
            //        // end for

            //        // increment index
            //        idx++;
            //        // convert p to 1d array
            //        Buffer.BlockCopy(p,0, p1, 0, p1.Length * sizeof(byte));
            //        // copy whole patch as a row to the indexed row of res;
            //        for (int i = 0; i < res.GetLength(0); i++)
            //            res[i, idx] = p1[i];
            //        // end for
            //        // copy the index of the particular iteration
            //        positions_out[0, idx] = y;
            //        positions_out[1, idx] = x;
            //        // every time idx reaches 1000, below executes. after that, idx is reset to 0
            //        // ------------------------ Prediction --------------------------------------
            //        if ( idx >= chunkSize - 1 )
            //        {
            //            // copy res to part
            //            Array.Copy(res, part, res.Length);
            //            for ( int i = 0; i<layer;i++)
            //            {
            //                for (int j = 0; j < part.GetLength(1);j++ )
            //                    part[part.GetLength(0) - layer + 1, j] = 1; 
            //                // end for
            //            } // end for
            //        } // end if
            //    }
            //}
            #endregion

            return counter;
        } // end method