Andersen Chen (aychen)

CS129 / Project 6 / Automated Panorama Stitching

We do automated panorama stitching by first collecting interest points on two corresponding images using the Harris Interest Point Detector. We then perform adaptive non-maximal suppression to cut down the number of interest points, extract normalized feature descriptors, and match corresponding features from the two images. We assume a projective transformation H that maps one image from another, and solve for the values in H. To make our process more robust to outliers, we use the RANSAC algorithm.

Results

The algorithm does well on all test cases and the new images. For a few images, the slightest misalignment is visible when there are sharp edges that get blurred.