Hongjian Zhang, 24261611
In this project, we are going to perform automatic stitching to turn images into a mosaic by doing the following:
1. Detecting corner features in an image;
2. Extracting a Feature Descriptor for each feature point;
3. Matching these feature descriptors between two images;
4. Use a robust method (RANSAC) to compute a homography
First we detect corners using Harris Corner Detector with a threshold of 1, and we end up with many interesting points. Then we perform Adaptive Non-Maximal Suppression to suppress and distribute interest points. We get a radius to represent the distance to nearest neighbor with a larger Harris intensity. Then, we sort the minimum radius for each point, and pick the largest 200 of them.
We extract axis-aligned 8x8 patches and sample these patches from the larger 40x40 window (spacing = 5 pixels) to get feature descriptors. To perform matching, we find pairs of features that look similar and are thus likely to be good matches. Then, we run RANSAC with 1000 iterations and 4 random choices. In each iteration, we sample 4 pair of matched correspondences to calculate the projective transformation. Then we transform image 1 accordingly. In the end, we pick the set of points with most inliers, and use its homography and blend with picture 2 to get the mosaic. Here we also compare the results obtained from manual selections and auto selections
The coolest thing I learned is how simple algorithms can complete seemingly very difficult and smart tasks, and sometimes they outperform humans.
