CC-SIFT: Exploiting Chromatic Contrast forWide-Baseline Matching

Abstract

Matching images taken from widely different viewpoints is still an open problem being extremely challenging. In this paper is presented a new method that manipulates effectively the color in order to improve the matching performances of the well-known SIFT operator for the wide-baseline case. Without exploiting additional information, the algorithm employs a new model that preserves efficiently the initial variations of the color and lightness. Instead of extracting feature points in the standard gray space, we take advantage of the redistributed chromatic information to increase the distinctness of the neighbor regions of the keypoints. No assumptions are made about the geometry or chromatic content of the initial images. The extensive experiments prove the utility and robustness of the new scheme that is able to find a considerable number of additional valid matches.