Image Matching Using Shift Features and Relocation Labeling in Dotnet

ABSTRACT:
An image matching algorithm is presented in this paper. A set of interest points known as SIFT features are computed for a pair of images. Every keypoint has a descriptor based on histogram of magnitude and direction of gradients. These descriptors are the primary input for the image correspondence algorithm. Initial probabilities are assigned for categories (probable matches) considering a feature point assignment to one of the category as a classification problem. Baye’s theorem is used for assigning initial probabilities. For selecting the neighbours for the left keypoint, a fixed number of pixels around the keypoint, considered as a window, are selected. The neighbours of the right keypoint are based on inspection of pair of images and the disparity range. The probabilistic estimates are iteratively improved by a relaxation labeling technique. The neighbour keypoints which will contribute to improve the probability is based on consistency property. The algorithm is effective for matching stereo image pair and these correspondences can be used as input for 3D reconstruction.