Project 3

Approach:

We followed the guidelines as closely as possible from the project page to implement the various steps needed to construct a panorama. Below are a summary of the steps taken to implement the various parts of Panorama.

Warp Sepherical:

Compute Euclidean coordinates Rotate Project to the z=1 plane Distort with k1 and k2

Feature Alignment:

One random feature match is chosen for each iteration of RANSAC Either estimate the translation or compute the homography Count inliers Update current largest inlier set After the RANSAC part is done, least squares fit is run to compute a transformation matrix

Blend Images:

Blend Images did not work as intended. It was supposed to: Compute the size of the final stitched image Resample each image to its final location and blend with its neighbors using a simple feathering function Crop the image to make the left and right edges seam perfectly

Results:

Since Blend Images did not work as intended, we were unable to construct a correct panorama. AlignPair using Homography was also untested. Here is the panorama we intended to use the program on:


panorama
(Created in Photoshop using the spherical distortion panorama photo merge.)