c# - EMGU CV SURF图像匹配

Question

我一直在使用 EMGU CV 库中的 SURF 特征检测示例。

到目前为止，它的效果非常好；我可以检测到 2 个给定图像之间的匹配对象，但我遇到了关于图像何时不匹配的问题。

我一直在寻求论坛的支持，但他们与我所在的位置相去甚远。谁知道哪些参数决定图像是否匹配。当我测试 2 张不匹配的图像时，代码仍然像存在匹配一样继续进行，即使没有匹配，也会在图像的随机位置绘制一条模糊的粗红线。

如果没有匹配项，我希望中断代码，不再继续。

附录:

      static void Run()
      {
          Image<Gray, Byte> modelImage = new Image<Gray, byte>("HatersGonnaHate.png");
         Image<Gray, Byte> observedImage = new Image<Gray, byte>("box_in_scene.png");
         Stopwatch watch;
         HomographyMatrix homography = null;

         SURFDetector surfCPU = new SURFDetector(500, false);

         VectorOfKeyPoint modelKeyPoints;
         VectorOfKeyPoint observedKeyPoints;
         Matrix<int> indices;
         Matrix<float> dist;
         Matrix<byte> mask;

         if (GpuInvoke.HasCuda)
         {
            GpuSURFDetector surfGPU = new GpuSURFDetector(surfCPU.SURFParams, 0.01f);
            using (GpuImage<Gray, Byte> gpuModelImage = new GpuImage<Gray, byte>(modelImage))
            //extract features from the object image
            using (GpuMat<float> gpuModelKeyPoints = surfGPU.DetectKeyPointsRaw(gpuModelImage, null))
            using (GpuMat<float> gpuModelDescriptors = surfGPU.ComputeDescriptorsRaw(gpuModelImage, null, gpuModelKeyPoints))
            using (GpuBruteForceMatcher matcher = new GpuBruteForceMatcher(GpuBruteForceMatcher.DistanceType.L2))
            {
               modelKeyPoints = new VectorOfKeyPoint();
               surfGPU.DownloadKeypoints(gpuModelKeyPoints, modelKeyPoints);
               watch = Stopwatch.StartNew();

               // extract features from the observed image
               using (GpuImage<Gray, Byte> gpuObservedImage = new GpuImage<Gray, byte>(observedImage))
               using (GpuMat<float> gpuObservedKeyPoints = surfGPU.DetectKeyPointsRaw(gpuObservedImage, null))
               using (GpuMat<float> gpuObservedDescriptors = surfGPU.ComputeDescriptorsRaw(gpuObservedImage, null, gpuObservedKeyPoints))
               using (GpuMat<int> gpuMatchIndices = new GpuMat<int>(gpuObservedDescriptors.Size.Height, 2, 1))
               using (GpuMat<float> gpuMatchDist = new GpuMat<float>(gpuMatchIndices.Size, 1))
               {
                  observedKeyPoints = new VectorOfKeyPoint();
                  surfGPU.DownloadKeypoints(gpuObservedKeyPoints, observedKeyPoints);

                  matcher.KnnMatch(gpuObservedDescriptors, gpuModelDescriptors, gpuMatchIndices, gpuMatchDist, 2, null);

                  indices = new Matrix<int>(gpuMatchIndices.Size);
                  dist = new Matrix<float>(indices.Size);
                  gpuMatchIndices.Download(indices);
                  gpuMatchDist.Download(dist);

                  mask = new Matrix<byte>(dist.Rows, 1);

                  mask.SetValue(255);

                  Features2DTracker.VoteForUniqueness(dist, 0.8, mask);

                  int nonZeroCount = CvInvoke.cvCountNonZero(mask);
                  if (nonZeroCount >= 4)
                  {
                     nonZeroCount = Features2DTracker.VoteForSizeAndOrientation(modelKeyPoints, observedKeyPoints, indices, mask, 1.5, 20);
                     if (nonZeroCount >= 4)
                        homography = Features2DTracker.GetHomographyMatrixFromMatchedFeatures(modelKeyPoints, observedKeyPoints, indices, mask, 3);
                  }

                  watch.Stop();
               }
            }
         }
         else
         {
            //extract features from the object image
            modelKeyPoints = surfCPU.DetectKeyPointsRaw(modelImage, null);
            //MKeyPoint[] kpts = modelKeyPoints.ToArray();
            Matrix<float> modelDescriptors = surfCPU.ComputeDescriptorsRaw(modelImage, null, modelKeyPoints);

            watch = Stopwatch.StartNew();

            // extract features from the observed image
            observedKeyPoints = surfCPU.DetectKeyPointsRaw(observedImage, null);
            Matrix<float> observedDescriptors = surfCPU.ComputeDescriptorsRaw(observedImage, null, observedKeyPoints);

            BruteForceMatcher matcher = new BruteForceMatcher(BruteForceMatcher.DistanceType.L2F32);
            matcher.Add(modelDescriptors);
            int k = 2;
            indices = new Matrix<int>(observedDescriptors.Rows, k);
            dist = new Matrix<float>(observedDescriptors.Rows, k);
            matcher.KnnMatch(observedDescriptors, indices, dist, k, null);

            mask = new Matrix<byte>(dist.Rows, 1);

            mask.SetValue(255);

            Features2DTracker.VoteForUniqueness(dist, 0.8, mask);

            int nonZeroCount = CvInvoke.cvCountNonZero(mask);
            if (nonZeroCount >= 4)
            {
               nonZeroCount = Features2DTracker.VoteForSizeAndOrientation(modelKeyPoints, observedKeyPoints, indices, mask, 1.5, 20);
               if (nonZeroCount >= 4)
                  homography = Features2DTracker.GetHomographyMatrixFromMatchedFeatures(modelKeyPoints, observedKeyPoints, indices, mask, 3);
            }

            watch.Stop();
         }

         //Draw the matched keypoints
        Image<Bgr, Byte> result = Features2DTracker.DrawMatches(modelImage, modelKeyPoints, observedImage, observedKeyPoints,
            indices, new Bgr(255, 255, 255), new Bgr(255, 255, 255), mask, Features2DTracker.KeypointDrawType.NOT_DRAW_SINGLE_POINTS);

         #region draw the projected region on the image
         if (homography != null)
         {  //draw a rectangle along the projected model
            Rectangle rect = modelImage.ROI;
            PointF[] pts = new PointF[] { 
               new PointF(rect.Left, rect.Bottom),
               new PointF(rect.Right, rect.Bottom),
               new PointF(rect.Right, rect.Top),
               new PointF(rect.Left, rect.Top)};
            homography.ProjectPoints(pts);

            result.DrawPolyline(Array.ConvertAll<PointF, Point>(pts, Point.Round), true, new Bgr(Color.Red), 5);
         }
         #endregion

         ImageViewer.Show(result, String.Format("Matched using {0} in {1} milliseconds", GpuInvoke.HasCuda ? "GPU" : "CPU", watch.ElapsedMilliseconds));
      }


   }

}

`

Answer 1

我不确定是否有适合所有图像序列或所有几何变形情况的方法。

我建议你计算PSNR在两个图像之间，并研究图像序列的容差阈值。