15-463 (Fall 2012) Computational Photography Project 6

1. Overview

In this project, functions to stitch images together and produce their mosaic automatically have been implemented. Given the correspondence points, a homography matrix is obtained. Using the homography matrix, we can warp/unwarp images to the same planar surface and blend them into one smooth wide-angle image.

2. Approach

For the first part of the project, the correspondence points are created by using cpselect tool in matlab. The homography matrix is calculated by solving the system of equations regarding homographic coordinates. Finally, we can process the projective transformation of images and interpolate their color values. The alpha blending technique is used to smoothen the overlapping areas.

For the second part of the project, the optimal correspondence points are automatically generated. The interst points are found using Harris Interest Point Detector and then suppressed by Adaptive Non-Maximal Suppression algorithm. For the spatially distributed interest points, the feature descriptors are obtained and normalized. The feature matching is conducted by Lowe's thresholding on the ratio between the first and the second nearest neighbors. Finally, RANSAC is used to eliminate outliers and to produce the best set of inliers. With the automatically generated correspondence points, we can use the image-stitching functions from the first part to make a photo mosaic.