Image Mosaic

Abstract

In this project we apply the Harris corner detector to find corners in two images, automatically find corresponding features, estimate a homography between them and warp one image into the coordinate system of the second one to produce a mosaic containing the union of all pixels in the two images. We also go further with this idea to create the panorama with three or more images.

Algorithm Design

• Corner Detection:
  apply the Harris corner detector to find corners and extract the features of each corner. Use non-maximum suppression to get a sparse set of corners.
• Find Correspondences:
  Compute normalized cross correlation (NCC) of the neighborhood at each corner. Set the threshold and find the best-matched points with higher NCC values.
• Estimate Homography:
  Use RANSAC algorithm to estimate the transformation matrix between two images.
• Create Panorama:
  Warp the image based on the estimated homography. Blend the warped image onto one panorama.