15-463

Sylvia Han (sylviah)

Auto-Stitching Photo Mosaics

Due: 11:59pm on Wednesday, November 21, 2012

Overview The goal of this project was to take images and perform a projection warp to create the same perspective for both images and then stitch them together. The second part of this project was to automate the stitching.

Approach For the first part of the project, the homography estimation is determined by using atleast four correspondence points for each of the images. These points are then used to solve a system of equations to determine the homography. We use the homography to warp one image to the perspective of another. We then stitch together the two images by using simple alpha blending. For the second part of the project, harris points were computed by the determining all the corners of an image. The amount of points were reduced using adapative non-maximal suppression. This smaller subset of points was then used to extract descriptors and create matchings between descriptors using Lowe's comparison of epsilon and the first nearest neighbor over the second nearest neighbor. The homography is then computed with these matched points. The homography is computed by RANSAC.

IMAGE RECTIFICATION: EXAMPLE 1

Original Rectified Cropped

IMAGE RECTIFICATION: EXAMPLE 2

Original Rectified Cropped

CHURCH

Part 1

Part 2

Red = Matched Feature Descriptors, Green = ANMS Points

COLLEGE OF FINE ARTS

Part 1

Part 2 - (Only two pictures)

Red = Matched Feature Descriptors, Green = ANMS Points

COLLEGE OF FINE ARTS CEILING

Part 1

Part 2

Red = Matched Feature Descriptors, Green = ANMS Points