public List <ImageRecord> QueryImage(string queryImagePath, out string messageToLog, SurfSettings surfSetting = null)
{
List <ImageRecord> rtnImageList = new List <ImageRecord>();
#region Diagnostic Region
Stopwatch sw = new Stopwatch();
long _loadingTime, _modelImageDectionlong, _queryingTime, _treeQuery, _loopTime = 0;
#endregion Diagnostic Region
#region Get KD-Tree Index
sw.Reset(); sw.Start();
//--------------Getting Indexed Records
SurfAccordDataSet surfAccordDataset;
bool isExist = CacheHelper.Get <SurfAccordDataSet>("SurfAccordDataSet", out surfAccordDataset);
if (!isExist)
{
string repoFileStoragePath = Path.Combine(DirectoryHelper.SaveDirectoryPath, "SurfAccordDataSet.bin");
if (!File.Exists(repoFileStoragePath))
{
string exMsg = string.Format("Can't get the Surf Index at {0}, please index first", repoFileStoragePath);
throw new FileNotFoundException(exMsg);
}
using (FileStream s = File.OpenRead(repoFileStoragePath))
{
//Polenter.Serialization.SharpSerializerBinarySettings bs =
// new Polenter.Serialization.SharpSerializerBinarySettings(Polenter.Serialization.BinarySerializationMode.SizeOptimized);
//Polenter.Serialization.SharpSerializer formatter = new Polenter.Serialization.SharpSerializer(bs);
//formatter.Serialize(surfAccordDataset, s);
System.Runtime.Serialization.Formatters.Binary.BinaryFormatter formatter
= new System.Runtime.Serialization.Formatters.Binary.BinaryFormatter();
surfAccordDataset = (SurfAccordDataSet)formatter.Deserialize(s);
s.Close();
}
CacheHelper.Add <SurfAccordDataSet>(surfAccordDataset, "SurfAccordDataSet");
}
if (surfAccordDataset == null)
{
throw new InvalidOperationException("Can't get the Surf Index, please index first");
}
sw.Stop();
_loadingTime = sw.ElapsedMilliseconds;
#endregion
#region Surf Dectator Region
double hessianThresh = 500;
double uniquenessThreshold = 0.8;
int goodMatchDistThreshold = 0;
if (surfSetting != null)
{
hessianThresh = surfSetting.HessianThresh.Value;
uniquenessThreshold = surfSetting.UniquenessThreshold.Value;
goodMatchDistThreshold = surfSetting.GoodMatchThreshold.Value;
}
float hessianThreshold2 = (float)hessianThresh / 1000000;
SpeededUpRobustFeaturesDetector surf = new SpeededUpRobustFeaturesDetector(hessianThreshold2);
#endregion Surf Dectator Region
#region Get Model Dectection and Validation
sw.Reset(); sw.Start();
List <SpeededUpRobustFeaturePoint> modelImageSurfPoints;
using (Bitmap modelImage = (Bitmap)Image.FromFile(queryImagePath))
{
modelImageSurfPoints = surf.ProcessImage(modelImage);
}
if (modelImageSurfPoints == null ||
modelImageSurfPoints.Count < 4)
{
throw new InvalidOperationException("Insuffucient interesting point in query image, try another query image");
}
sw.Stop();
_modelImageDectionlong = sw.ElapsedMilliseconds;
#endregion
#region Search Images
sw.Reset(); sw.Start();
//------------Search Images
Accord.MachineLearning.Structures.KDTree <int> tree = surfAccordDataset.IndexedTree;
double[][] listofQueryDescriptors = modelImageSurfPoints.Select(ft => ft.Descriptor).ToArray();
double[] myscores = new double[listofQueryDescriptors.Length];
int[] labels = Enumerable.Repeat(-1, listofQueryDescriptors.Length).ToArray();
for (int i = 0; i < listofQueryDescriptors.Length; i++)
{
KDTreeNodeCollection <int> neighbors = tree.ApproximateNearest(listofQueryDescriptors[i], 2, 90d);
//KDTreeNodeCollection<int> neighbors = tree.Nearest(listofQueryDescriptors[i], uniquenessThreshold, 2);
Dictionary <int, double> keyValueStore = new Dictionary <int, double>();
double similarityDist = 0;
foreach (KDTreeNodeDistance <int> point in neighbors)
{
int label = point.Node.Value;
double d = point.Distance;
// Convert to similarity measure
if (keyValueStore.ContainsKey(label))
{
similarityDist = keyValueStore[label];
similarityDist += 1.0 / (1.0 + d);
keyValueStore[label] = similarityDist;
}
else
{
similarityDist = 1.0 / (1.0 + d);
keyValueStore.Add(label, similarityDist);
}
}
if (keyValueStore.Count > 0)
{
int maxIndex = keyValueStore.Aggregate((l, r) => l.Value > r.Value ? l : r).Key;
labels[i] = maxIndex;
double sumOfAllValues = keyValueStore.Values.Sum();
myscores[i] = keyValueStore[maxIndex] / sumOfAllValues;
}
}
sw.Stop();
_queryingTime = sw.ElapsedMilliseconds;
sw.Reset(); sw.Start();
List <SURFRecord1> listOfSurfImages = surfAccordDataset.SurfImageIndexRecord;
//----------Create Interval Tree from ImageMetaData
IntervalTreeLib.IntervalTree <SURFRecord1, int> intervalTree;
bool isTreeExist = CacheHelper.Get <IntervalTreeLib.IntervalTree <SURFRecord1, int> >("SurfAccordIntervalTree", out intervalTree);
if (!isTreeExist)
{
intervalTree = new IntervalTreeLib.IntervalTree <SURFRecord1, int>();
foreach (var record in listOfSurfImages)
{
intervalTree.AddInterval(record.IndexStart, record.IndexEnd, record);
}
CacheHelper.Add <IntervalTreeLib.IntervalTree <SURFRecord1, int> >(intervalTree, "SurfAccordIntervalTree");
}
//--------------Matching Target image similarity
for (int i = 0; i < listofQueryDescriptors.Length; i++)
{
int rowNum = labels[i];
if (rowNum == -1)
{
continue;
}
double dist = myscores[i];
SURFRecord1 rec = intervalTree.Get(rowNum, IntervalTreeLib.StubMode.ContainsStartThenEnd).FirstOrDefault();
rec.Distance++;
}
sw.Stop();
_loopTime = sw.ElapsedMilliseconds;
#endregion
string msg = String.Format("Loading: {0}, Model detection: {1}, Querying: {2}, Looping: {3}",
_loadingTime, _modelImageDectionlong, _queryingTime, _loopTime);
messageToLog = msg;
rtnImageList = listOfSurfImages.Where(rec => rec.Distance > goodMatchDistThreshold)
.OrderByDescending(rec => rec.Distance)
.ToList <ImageRecord>();
return(rtnImageList);
}
public List<ImageRecord> QueryImage(string queryImagePath, out string messageToLog, SurfSettings surfSetting = null)
{
List<ImageRecord> rtnImageList = new List<ImageRecord>();
#region Diagnostic Region
Stopwatch sw = new Stopwatch();
long _loadingTime, _modelImageDectionlong, _queryingTime, _treeQuery, _loopTime = 0;
#endregion Diagnostic Region
#region Get KD-Tree Index
sw.Reset(); sw.Start();
//--------------Getting Indexed Records
SurfAccordDataSet surfAccordDataset;
bool isExist= CacheHelper.Get<SurfAccordDataSet>("SurfAccordDataSet", out surfAccordDataset);
if (!isExist)
{
string repoFileStoragePath = Path.Combine(DirectoryHelper.SaveDirectoryPath, "SurfAccordDataSet.bin");
if (!File.Exists(repoFileStoragePath))
{
string exMsg = string.Format("Can't get the Surf Index at {0}, please index first", repoFileStoragePath);
throw new FileNotFoundException(exMsg);
}
using (FileStream s = File.OpenRead(repoFileStoragePath))
{
//Polenter.Serialization.SharpSerializerBinarySettings bs =
// new Polenter.Serialization.SharpSerializerBinarySettings(Polenter.Serialization.BinarySerializationMode.SizeOptimized);
//Polenter.Serialization.SharpSerializer formatter = new Polenter.Serialization.SharpSerializer(bs);
//formatter.Serialize(surfAccordDataset, s);
System.Runtime.Serialization.Formatters.Binary.BinaryFormatter formatter
= new System.Runtime.Serialization.Formatters.Binary.BinaryFormatter();
surfAccordDataset = (SurfAccordDataSet)formatter.Deserialize(s);
s.Close();
}
CacheHelper.Add<SurfAccordDataSet>(surfAccordDataset, "SurfAccordDataSet");
}
if (surfAccordDataset == null)
throw new InvalidOperationException("Can't get the Surf Index, please index first");
sw.Stop();
_loadingTime = sw.ElapsedMilliseconds;
#endregion
#region Surf Dectator Region
double hessianThresh = 500;
double uniquenessThreshold = 0.8;
int goodMatchDistThreshold = 0;
if (surfSetting != null)
{
hessianThresh = surfSetting.HessianThresh.Value;
uniquenessThreshold = surfSetting.UniquenessThreshold.Value;
goodMatchDistThreshold = surfSetting.GoodMatchThreshold.Value;
}
float hessianThreshold2 = (float)hessianThresh / 1000000;
SpeededUpRobustFeaturesDetector surf = new SpeededUpRobustFeaturesDetector(hessianThreshold2);
#endregion Surf Dectator Region
#region Get Model Dectection and Validation
sw.Reset(); sw.Start();
List<SpeededUpRobustFeaturePoint> modelImageSurfPoints;
using (Bitmap modelImage = (Bitmap)Image.FromFile(queryImagePath))
{
modelImageSurfPoints = surf.ProcessImage(modelImage);
}
if (modelImageSurfPoints == null
|| modelImageSurfPoints.Count < 4)
{
throw new InvalidOperationException("Insuffucient interesting point in query image, try another query image");
}
sw.Stop();
_modelImageDectionlong = sw.ElapsedMilliseconds;
#endregion
#region Search Images
sw.Reset(); sw.Start();
//------------Search Images
Accord.MachineLearning.Structures.KDTree<int> tree = surfAccordDataset.IndexedTree;
double[][] listofQueryDescriptors = modelImageSurfPoints.Select(ft => ft.Descriptor).ToArray();
double[] myscores = new double[listofQueryDescriptors.Length];
int[] labels = Enumerable.Repeat(-1, listofQueryDescriptors.Length).ToArray();
for (int i = 0; i < listofQueryDescriptors.Length; i++)
{
KDTreeNodeCollection<int> neighbors = tree.ApproximateNearest(listofQueryDescriptors[i],2, 90d);
//KDTreeNodeCollection<int> neighbors = tree.Nearest(listofQueryDescriptors[i], uniquenessThreshold, 2);
Dictionary<int, double> keyValueStore = new Dictionary<int, double>();
double similarityDist = 0;
foreach (KDTreeNodeDistance<int> point in neighbors)
{
int label = point.Node.Value;
double d = point.Distance;
// Convert to similarity measure
if (keyValueStore.ContainsKey(label))
{
similarityDist = keyValueStore[label];
similarityDist += 1.0 / (1.0 + d);
keyValueStore[label] = similarityDist;
}
else
{
similarityDist = 1.0 / (1.0 + d);
keyValueStore.Add(label, similarityDist);
}
}
if (keyValueStore.Count > 0)
{
int maxIndex = keyValueStore.Aggregate((l, r) => l.Value > r.Value ? l : r).Key;
labels[i] = maxIndex;
double sumOfAllValues = keyValueStore.Values.Sum();
myscores[i] = keyValueStore[maxIndex] / sumOfAllValues;
}
}
sw.Stop();
_queryingTime = sw.ElapsedMilliseconds;
sw.Reset(); sw.Start();
List<SURFRecord1> listOfSurfImages = surfAccordDataset.SurfImageIndexRecord;
//----------Create Interval Tree from ImageMetaData
IntervalTreeLib.IntervalTree<SURFRecord1, int> intervalTree;
bool isTreeExist = CacheHelper.Get<IntervalTreeLib.IntervalTree<SURFRecord1, int>>("SurfAccordIntervalTree", out intervalTree);
if (!isTreeExist)
{
intervalTree = new IntervalTreeLib.IntervalTree<SURFRecord1, int>();
foreach (var record in listOfSurfImages)
{
intervalTree.AddInterval(record.IndexStart, record.IndexEnd, record);
}
CacheHelper.Add<IntervalTreeLib.IntervalTree<SURFRecord1, int>>(intervalTree, "SurfAccordIntervalTree");
}
//--------------Matching Target image similarity
for (int i = 0; i < listofQueryDescriptors.Length; i++)
{
int rowNum = labels[i];
if (rowNum == -1) continue;
double dist = myscores[i];
SURFRecord1 rec = intervalTree.Get(rowNum, IntervalTreeLib.StubMode.ContainsStartThenEnd).FirstOrDefault();
rec.Distance++;
}
sw.Stop();
_loopTime = sw.ElapsedMilliseconds;
#endregion
string msg = String.Format("Loading: {0}, Model detection: {1}, Querying: {2}, Looping: {3}",
_loadingTime, _modelImageDectionlong, _queryingTime, _loopTime);
messageToLog = msg;
rtnImageList = listOfSurfImages.Where(rec => rec.Distance > goodMatchDistThreshold)
.OrderByDescending(rec => rec.Distance)
.ToList<ImageRecord>();
return rtnImageList;
}
