+
+
+
=
Here is my panorama stitcher applied to 4 images of Gates:
+++=
How is this done? First, the Harris corner algorithm is applied to find points of interest in the image.
Then, 8x8 patches are extracted by scaling down the 40x40 region around each Harris corner. Then each point in the first image is matched to a point in the second image by comparing patches.
Most of these matches are bad, so only those matches where the best match is significantly better than the second best match are shown.
Finally, the RANSAC algorithm, in which 4 points are randomly chosen from one image to create a transformation and said transformation is then applied to every point in the first image. This is repeated 1000 times, and the transformation that aligns the most transformed points to points on the second image is chosen. These alignable points are returned.
Using these aligned point sets, the images are merged.
