public void Panorama_Example1()
{
Accord.Math.Tools.SetupGenerator(0);
// Let's start with two pictures that have been
// taken from slightly different points of view:
//
Bitmap img1 = Resources.dc_left;
Bitmap img2 = Resources.dc_right;
// Those pictures are shown below:
// ImageBox.Show(img1, PictureBoxSizeMode.Zoom, 640, 480);
// ImageBox.Show(img2, PictureBoxSizeMode.Zoom, 640, 480);
// Step 1: Detect feature points using Surf Corners Detector
var surf = new SpeededUpRobustFeaturesDetector();
var points1 = surf.ProcessImage(img1);
var points2 = surf.ProcessImage(img2);
// Step 2: Match feature points using a k-NN
var matcher = new KNearestNeighborMatching(5);
var matches = matcher.Match(points1, points2);
// Step 3: Create the matrix using a robust estimator
var ransac = new RansacHomographyEstimator(0.001, 0.99);
MatrixH homographyMatrix = ransac.Estimate(matches);
Assert.AreEqual(1.13583624f, homographyMatrix.Elements[0], 1e-5);
Assert.AreEqual(-0.0229569562f, homographyMatrix.Elements[1], 1e-5);
Assert.AreEqual(-255.243988f, homographyMatrix.Elements[2], 1e-2);
Assert.AreEqual(0.080111593f, homographyMatrix.Elements[3], 1e-5);
Assert.AreEqual(1.11404252f, homographyMatrix.Elements[4], 1e-5);
Assert.AreEqual(-167.362167f, homographyMatrix.Elements[5], 1e-2);
Assert.AreEqual(0.00011207442f, homographyMatrix.Elements[6], 1e-5);
Assert.AreEqual(0.0000529394056f, homographyMatrix.Elements[7], 1e-5);
Assert.AreEqual(8, homographyMatrix.Elements.Length);
// Step 4: Project and blend using the homography
Blend blend = new Blend(homographyMatrix, img1);
// Compute the blending algorithm
Bitmap result = blend.Apply(img2);
// Show on screen
// ImageBox.Show(result, PictureBoxSizeMode.Zoom, 640, 480);
#pragma warning disable 618
double[,] expected = Properties.Resources.blend_result.ToDoubleMatrix(0);
double[,] actual = result.ToDoubleMatrix(0);
Assert.IsTrue(Matrix.IsEqual(expected, actual, 0.1));
#pragma warning restore 618
}
public void Example1()
{
// Let's start with two pictures that have been
// taken from slightly different points of view:
//
Bitmap img1 = Resources.dc_left;
Bitmap img2 = Resources.dc_right;
// Those pictures are shown below:
ImageBox.Show(img1, PictureBoxSizeMode.Zoom, 640, 480);
ImageBox.Show(img2, PictureBoxSizeMode.Zoom, 640, 480);
// Step 1: Detect feature points using Surf Corners Detector
var surf = new SpeededUpRobustFeaturesDetector();
var points1 = surf.ProcessImage(img1);
var points2 = surf.ProcessImage(img2);
// Step 2: Match feature points using a k-NN
var matcher = new KNearestNeighborMatching(5);
var matches = matcher.Match(points1, points2);
// Step 3: Create the matrix using a robust estimator
var ransac = new RansacHomographyEstimator(0.001, 0.99);
MatrixH homographyMatrix = ransac.Estimate(matches);
// Step 4: Project and blend using the homography
Blend blend = new Blend(homographyMatrix, img1);
// Compute the blending algorithm
Bitmap result = blend.Apply(img2);
// Show on screen
ImageBox.Show(result, PictureBoxSizeMode.Zoom, 640, 480);
}
private void btnRansac_Click(object sender, EventArgs e)
{
if (correlationPoints1 == null)
{
MessageBox.Show("Please, click Nearest Neighbor button first! :-)");
return;
}
if (correlationPoints1.Length < 4 || correlationPoints2.Length < 4)
{
MessageBox.Show("Insufficient points to attempt a fit.");
return;
}
// Step 3: Create the homography matrix using a robust estimator
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
homography = ransac.Estimate(correlationPoints1, correlationPoints2);
// Plot RANSAC results against correlation results
IntPoint[] inliers1 = correlationPoints1.Submatrix(ransac.Inliers);
IntPoint[] inliers2 = correlationPoints2.Submatrix(ransac.Inliers);
// Concatenate the two images in a single image (just to show on screen)
Concatenate concat = new Concatenate(img1);
Bitmap img3 = concat.Apply(img2);
// Show the marked correlations in the concatenated image
PairsMarker pairs = new PairsMarker(
inliers1, // Add image1's width to the X points to show the markings correctly
inliers2.Apply(p => new IntPoint(p.X + img1.Width, p.Y)));
pictureBox.Image = pairs.Apply(img3);
}
/// <summary>
/// Matches detected keypoints.
/// </summary>
private void MatchKeypoints()
{
// matching keypoints step needs at least two images
if (input_images.Count < 2) return;
// using RANSAC homography estimator
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
CorrelationMatching matcher = new CorrelationMatching(9);
Bitmap mergedImage = new Bitmap(input_images.Count * input_images[0].Width, input_images[0].Height); //it is assumed the images are of same size!
Graphics graphics = Graphics.FromImage(mergedImage);
IntPoint[][] matches;
int cumulativeWidth = 0;
int keypoints_cntr = 0;
// draw all images
for (int i = 0; i < input_images.Count; i++) {
graphics.DrawImage(input_images[i], cumulativeWidth, 0, input_images[i].Width, input_images[i].Height);
cumulativeWidth += input_images[i].Width;
}
// iterate through all images
for (int i = 0; i < input_images.Count; i++)
{
if (i != input_images.Count - 1)
{
// match detected keypoints with maximum cross-correlation algorithm
matches = matcher.Match(new Bitmap(input_images[i]), new Bitmap(input_images[i + 1]), keypoints[keypoints_cntr].ToArray(), keypoints[keypoints_cntr + 1].ToArray());
IntPoint[] correlationPoints1 = matches[0];
IntPoint[] correlationPoints2 = matches[1];
// Plot RANSAC results against correlation results
homography = ransac.Estimate(correlationPoints1, correlationPoints2);
// take only inliers - good matches
IntPoint[] inliers1 = correlationPoints1.Submatrix(ransac.Inliers);
IntPoint[] inliers2 = correlationPoints2.Submatrix(ransac.Inliers);
PairsMarker pairs = new PairsMarker(correlationPoints1, correlationPoints2.Apply(p => new IntPoint(p.X + input_images[i].Width, p.Y)));
// draw all matching pairs
for (int j = 0; j < pairs.Points1.Count(); j++) {
if (i % 2 == 0) graphics.DrawLine(new Pen(Color.Red, 1.5f), new System.Drawing.Point(correlationPoints1[j].X + i * input_images[i].Width, correlationPoints1[j].Y), new System.Drawing.Point(correlationPoints2[j].X + (i + 1) * input_images[i].Width, correlationPoints2[j].Y));
}
// draw inliers
for (int j = 0; j < inliers1.Count(); j++) {
if (i % 2 == 0) graphics.DrawLine(new Pen(Color.GreenYellow, 1.5f), new System.Drawing.Point(inliers1[j].X + i * input_images[i].Width, inliers1[j].Y), new System.Drawing.Point(inliers2[j].X + (i + 1) * input_images[i].Width, inliers2[j].Y));
else graphics.DrawLine(new Pen(Color.Blue, 1.5f), new System.Drawing.Point(inliers1[j].X + i * input_images[i].Width, inliers1[j].Y), new System.Drawing.Point(inliers2[j].X + (i + 1) * input_images[i].Width, inliers2[j].Y));
}
keypoints_cntr += 2;
}
}
ShowThumbnail(mergedImage, true, PanoramaPhase.MatchKeypoints);
}
public void EstimateTest()
{
Point[] points1 =
{
new Point(86, 3),
new Point(262, 7),
new Point(72, 12),
new Point(233, 14),
new Point(222, 16),
new Point(242, 19),
new Point(174, 21),
new Point(199, 22),
new Point(210, 23),
new Point(245, 27),
new Point(223, 28),
new Point(171, 29),
new Point(213, 32),
new Point(206, 34),
new Point(158, 36),
new Point(215, 36),
new Point(194, 40),
new Point(155, 43),
new Point(390, 145),
new Point(325, 151),
new Point(430, 165),
new Point(569, 166),
new Point(548, 171),
new Point(486, 172),
new Point(585, 174),
new Point(513, 175),
new Point(581, 178)
};
//Points 2
Point[] points2 =
{
new Point(94, 3),
new Point(129, 10),
new Point(135, 6),
new Point(100, 16),
new Point(88, 18),
new Point(109, 22),
new Point(35, 23),
new Point(63, 24),
new Point(75, 25),
new Point(112, 30),
new Point(89, 31),
new Point(32, 31),
new Point(78, 35),
new Point(70, 37),
new Point(19, 38),
new Point(80, 39),
new Point(58, 43),
new Point(15, 46),
new Point(259, 151),
new Point(194, 158),
new Point(299, 171),
new Point(433, 171),
new Point(414, 176),
new Point(354, 177),
new Point(449, 178),
new Point(380, 180),
new Point(445, 183)
};
double[,] expected =
{
{ 0.60628712500923021, 0.00059969215221173516, -85.656775800903588 },
{ 0.010863088422024825, 0.58853684011367191, -1.6919055825149059 },
{ 0.000088084825486304467, 0.000063754043404499572, 0.53717560168513312 }
};
expected = (double[,])(new MatrixH(expected));
// Set a fixed seed to transform RANSAC into a deterministic algorithm
Accord.Math.Random.Generator.Seed = 0;
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
double[,] actual = (double[,])ransac.Estimate(points1, points2);
for (int i = 0; i < 3; i++)
for (int j = 0; j < 3; j++)
Assert.AreEqual(actual[i, j], expected[i, j], 0.001);
}
private Bitmap CompareAndDrawImage(Bitmap modelImage, Bitmap observedImage, SurfSettings setting)
{
Stopwatch watch1 = new Stopwatch();
Stopwatch watch2 = new Stopwatch();
Bitmap returnBitmap;
watch2.Start();
watch1.Reset(); watch1.Start();
double hessianThreshold = setting.HessianThresh.HasValue ? setting.HessianThresh.Value : 500;
float hessianThreshold2 = (float)hessianThreshold / 1000000;
Debug.WriteLine("hessianThreshold2: {0}", hessianThreshold2);
SpeededUpRobustFeaturesDetector surf = new SpeededUpRobustFeaturesDetector(hessianThreshold2);
List<SpeededUpRobustFeaturePoint> surfPoints1 = surf.ProcessImage(modelImage);
List<SpeededUpRobustFeaturePoint> surfPoints2 = surf.ProcessImage(observedImage);
Debug.WriteLine("Surf points count: {0}", surfPoints1.Count);
Debug.WriteLine("Surf points count: {0}", surfPoints2.Count);
//long memoryFootprint = MemorySize.GetBlobSizeinKb(surfPoints2);
//Debug.WriteLine("Surf extractor: {0} kb", memoryFootprint);
watch1.Stop();
Debug.WriteLine("Surf Detection tooked {0} ms", watch1.ElapsedMilliseconds);
watch1.Reset(); watch1.Start();
// Show the marked points in the original images
Bitmap img1mark = new FeaturesMarker(surfPoints1, 2).Apply(modelImage);
Bitmap img2mark = new FeaturesMarker(surfPoints2, 2).Apply(observedImage);
// Concatenate the two images together in a single image (just to show on screen)
Concatenate concatenate = new Concatenate(img1mark);
returnBitmap = concatenate.Apply(img2mark);
watch1.Stop();
Debug.WriteLine("Surf point plotting tooked {0} ms", watch1.ElapsedMilliseconds);
//watch1.Reset(); watch1.Start();
//List<IntPoint>[] coretionalMatches = getMatches(surfPoints1, surfPoints2);
//watch1.Stop();
//Debug.WriteLine("Correctional Match tooked {0} ms", watch1.ElapsedMilliseconds);
//// Get the two sets of points
//IntPoint[] correlationPoints11 = coretionalMatches[0].ToArray();
//IntPoint[] correlationPoints22 = coretionalMatches[1].ToArray();
//Debug.WriteLine("Correclation points count: {0}", correlationPoints11.Length);
//Debug.WriteLine("Correclation points count: {0}", correlationPoints22.Length);
Debug.WriteLine("Threshold: {0}", setting.UniquenessThreshold.Value);
watch1.Reset(); watch1.Start();
// Step 2: Match feature points using a k-NN
KNearestNeighborMatching matcher = new KNearestNeighborMatching(2);
matcher.Threshold = setting.UniquenessThreshold.Value;
IntPoint[][] matches = matcher.Match(surfPoints1, surfPoints2);
watch1.Stop();
Debug.WriteLine("Knn Match tooked {0} ms", watch1.ElapsedMilliseconds);
// Get the two sets of points
IntPoint[] correlationPoints1 = matches[0];
IntPoint[] correlationPoints2 = matches[1];
Debug.WriteLine("Knn points count: {0}", correlationPoints1.Length);
Debug.WriteLine("Knn points count: {0}", correlationPoints2.Length);
//watch1.Reset(); watch1.Start();
//// Show the marked correlations in the concatenated image
//PairsMarker pairs = new PairsMarker(
// correlationPoints1, // Add image1's width to the X points to show the markings correctly
// correlationPoints2.Apply(p => new IntPoint(p.X + modelImage.Width, p.Y)), Color.Blue);
//returnBitmap = pairs.Apply(returnBitmap);
//watch1.Stop();
//Debug.WriteLine("Match pair marking tooked {0} ms", watch1.ElapsedMilliseconds);
if (correlationPoints1.Length < 4 || correlationPoints2.Length < 4)
{
MessageBox.Show("Insufficient points to attempt a fit.");
return null;
}
watch1.Reset(); watch1.Start();
// Step 3: Create the homography matrix using a robust estimator
//RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
MatrixH homography = ransac.Estimate(correlationPoints1, correlationPoints2);
watch1.Stop();
Debug.WriteLine("Ransac tooked {0} ms", watch1.ElapsedMilliseconds);
watch1.Reset(); watch1.Start();
// Plot RANSAC results against correlation results
IntPoint[] inliers1 = correlationPoints1.Submatrix(ransac.Inliers);
IntPoint[] inliers2 = correlationPoints2.Submatrix(ransac.Inliers);
watch1.Stop();
Debug.WriteLine("Ransac SubMatrix {0} ms", watch1.ElapsedMilliseconds);
Debug.WriteLine("Ransac points count: {0}", inliers1.Length);
Debug.WriteLine("Ransac points count: {0}", inliers2.Length);
watch1.Reset(); watch1.Start();
PairsMarker inlierPairs = new PairsMarker(
inliers1, // Add image1's width to the X points to show the markings correctly
inliers2.Apply(p => new IntPoint(p.X + modelImage.Width, p.Y)), Color.Red);
returnBitmap = inlierPairs.Apply(returnBitmap);
watch1.Stop();
Debug.WriteLine("Ransac plotting tooked {0} ms", watch1.ElapsedMilliseconds);
watch2.Stop();
return returnBitmap;
}
private bool GenerateKeypointMatches(Bitmap target, Bitmap frame, Rectangle prevTarget)
{
HarrisCornersDetector harrisDetector = new HarrisCornersDetector(HarrisCornerMeasure.Harris, 1000f, 1f, 2);
targetKeyPoints=harrisDetector.ProcessImage(target).ToArray();
frameKeyPoints=harrisDetector.ProcessImage(frame).ToArray();
//Console.WriteLine("tr={0} fr={1}", targetKeyPoints.Length, frameKeyPoints.Length);
if (targetKeyPoints.Length==0||frameKeyPoints.Length==0)
{
return false;
}
CorrelationMatching matcher=new CorrelationMatching(15);
IntPoint[][] matches=matcher.Match(target, frame, targetKeyPoints, frameKeyPoints);
targetMacthes=matches[0];
frameMatches=matches[1];
if (targetMacthes.Length<4||frameMatches.Length<4)
{
return false;
}
RansacHomographyEstimator ransac=new RansacHomographyEstimator(0.1, 0.50);
MatrixH estiment = new MatrixH();
try
{
estiment = ransac.Estimate(targetMacthes, frameMatches);
}
catch
{
return false;
}
IntPoint[] targetGoodMatch=targetMacthes.Submatrix(ransac.Inliers);
IntPoint[] frameGoodMatch=frameMatches.Submatrix(ransac.Inliers);
CalculatePosChange(targetGoodMatch, frameGoodMatch, prevTarget);
return true;
}
private void BtnRansac_OnClick(object sender, RoutedEventArgs e)
{
// Step 3: Create the homography matrix using a robust estimator
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
homography = ransac.Estimate(correlationPoints1, correlationPoints2);
// Plot RANSAC results against correlation results
IntPoint[] inliers1 = correlationPoints1.Submatrix(ransac.Inliers);
IntPoint[] inliers2 = correlationPoints2.Submatrix(ransac.Inliers);
// Concatenate the two images in a single image (just to show on screen)
Concatenate concat = new Concatenate(img1);
Bitmap img3 = concat.Apply(img2);
// Show the marked correlations in the concatenated image
PairsMarker pairs = new PairsMarker(
inliers1, // Add image1's width to the X points to show the markings correctly
inliers2.Apply(p => new IntPoint(p.X + img1.Width, p.Y)));
PictureBox.Source = pairs.Apply(img3);
}
public void Panorama_Example1()
{
Accord.Math.Tools.SetupGenerator(0);
// Let's start with two pictures that have been
// taken from slightly different points of view:
//
Bitmap img1 = Resources.dc_left;
Bitmap img2 = Resources.dc_right;
// Those pictures are shown below:
// ImageBox.Show(img1, PictureBoxSizeMode.Zoom, 640, 480);
// ImageBox.Show(img2, PictureBoxSizeMode.Zoom, 640, 480);
// Step 1: Detect feature points using Surf Corners Detector
var surf = new SpeededUpRobustFeaturesDetector();
var points1 = surf.ProcessImage(img1);
var points2 = surf.ProcessImage(img2);
// Step 2: Match feature points using a k-NN
var matcher = new KNearestNeighborMatching(5);
var matches = matcher.Match(points1, points2);
// Step 3: Create the matrix using a robust estimator
var ransac = new RansacHomographyEstimator(0.001, 0.99);
MatrixH homographyMatrix = ransac.Estimate(matches);
Assert.AreEqual(1.15707409, homographyMatrix.Elements[0], 1e-5);
Assert.AreEqual(-0.0233834628, homographyMatrix.Elements[1], 1e-5);
Assert.AreEqual(-261.8217, homographyMatrix.Elements[2], 1e-2);
Assert.AreEqual(0.08801343, homographyMatrix.Elements[3], 1e-5);
Assert.AreEqual(1.12451434, homographyMatrix.Elements[4], 1e-5);
Assert.AreEqual(-171.191208, homographyMatrix.Elements[5], 1e-2);
Assert.AreEqual(0.000127789128, homographyMatrix.Elements[6], 1e-5);
Assert.AreEqual(0.00006173445, homographyMatrix.Elements[7], 1e-5);
Assert.AreEqual(8, homographyMatrix.Elements.Length);
// Step 4: Project and blend using the homography
Blend blend = new Blend(homographyMatrix, img1);
// Compute the blending algorithm
Bitmap result = blend.Apply(img2);
// Show on screen
// ImageBox.Show(result, PictureBoxSizeMode.Zoom, 640, 480);
//result.Save(@"C:\Projects\Accord.NET\net35.png", ImageFormat.Png);
#if NET35
Bitmap image = Properties.Resources.blend_net35;
#else
Bitmap image = Properties.Resources.blend_net45;
#endif
#pragma warning disable 618
double[,] expected = image.ToDoubleMatrix(0);
double[,] actual = result.ToDoubleMatrix(0);
Assert.IsTrue(Matrix.IsEqual(expected, actual, 0.1));
#pragma warning restore 618
}
private void button2_Click(object sender, EventArgs e)
{
// Step 3: Create the homography matrix using a robust estimator
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
homography = ransac.Estimate(correlationPoints1, correlationPoints2);
// Plot RANSAC results against correlation results
IntPoint[] inliers1 = correlationPoints1.Submatrix(ransac.Inliers);
IntPoint[] inliers2 = correlationPoints2.Submatrix(ransac.Inliers);
// Concatenate the two images in a single image (just to show on screen)
Concatenate concat = new Concatenate(img1);
Bitmap img3 = concat.Apply(img2);
// Show the marked correlations in the concatenated image
PairsMarker pairs = new PairsMarker(
inliers1, // Add image1's width to the X points to show the markings correctly
inliers2.Apply(p => new IntPoint(p.X + img1.Width, p.Y)));
pictureBox1.Image = pairs.Apply(img3);
numA.Value = (decimal)homography.Elements[0];
numB.Value = (decimal)homography.Elements[1];
numC.Value = (decimal)homography.Elements[2];
numD.Value = (decimal)homography.Elements[3];
numE.Value = (decimal)homography.Elements[4];
numF.Value = (decimal)homography.Elements[5];
numG.Value = (decimal)homography.Elements[6];
numH.Value = (decimal)homography.Elements[7];
}
// PANORAMIC STICHING CODE
// Accord.NET and AForge.NET frameworks and code examples provided from
// http://www.codeproject.com/KB/recipes/automatic_panoramas.aspx
private void panoramicStitchingToolStripMenuItem_Click(object sender, EventArgs e)
{
// Save a copy of the current image, ask user to open another image to merge with
Bitmap img2 = img;
openToolStripMenuItem_Click(sender, e);
// Check whether the current loaded image is different
// (If the user cancelled the open image operation, the current image in the viewer
// is still the same image object)
if (img2 != img)
{
Bitmap img1 = img;
AForge.IntPoint[] harrisPoints1;
AForge.IntPoint[] harrisPoints2;
AForge.IntPoint[] correlationPoints1;
AForge.IntPoint[] correlationPoints2;
MatrixH homography;
// Use Harris Corners Detector to find points of interest
HarrisCornersDetector harris = new HarrisCornersDetector(0.04f, 1000f);
harrisPoints1 = harris.ProcessImage(img1).ToArray();
harrisPoints2 = harris.ProcessImage(img2).ToArray();
// This check fixes an out of bounds exception generated by matcher.Match() below when a
// monocolour image (0 harris points) is stitched with a non-monocolour image
if (harrisPoints1.Length == 0 || harrisPoints2.Length == 0)
{
MessageBox.Show("Panoramic stitching cannot continue because at least one of the images does not contain any Harris points.", "Warning");
}
else
{
// Match detected points using correlation
CorrelationMatching matcher = new CorrelationMatching(9);
AForge.IntPoint[][] matches = matcher.Match(img1, img2, harrisPoints1, harrisPoints2);
// Separate the two arrays
correlationPoints1 = matches[0];
correlationPoints2 = matches[1];
// Find homography matrix using RANSAC algorithm
RansacHomographyEstimator ransac = new RansacHomographyEstimator(0.001, 0.99);
// This check is to handle the ransac.Estimate() function
// which throws an exception if the array parameters do not contain at least 4 elements
if (correlationPoints1.Length < 4 || correlationPoints2.Length < 4)
{
MessageBox.Show("Panoramic stitching cannot continue because at least one of the images does not contain at least 4 correlation points.", "Warning");
}
else
{
homography = ransac.Estimate(correlationPoints1, correlationPoints2);
// Merge the images
Blend blend = new Blend(homography, img1);
img = blend.Apply(img2);
//save the image properly and resize main form
origImg.Dispose();
origImg = new Bitmap(img);
pictureBox.Image = img;
mainForm.ActiveForm.Width = img.Width + widthPad;
mainForm.ActiveForm.Height = img.Height + heightPad;
}
}
}
}