Robust Feature Matching Across Widely Separated Color Images

We present a novel method for feature matching across widely separated color images. The proposed approach is robust and can support various correspondence based algorithms e.g. the recovery of epipolar geometry. Our algorithm extends an existing gray-scale corner detector to color. The feature matching algorithm robustly segments the area around the feature into significant color regions using the mean shift mode estimator. The recovered data structures are matched under all possible rotations and the best rotation and its corresponding matches are selected. The results of the matching algorithm are used for recovery of the epipolar geometry from wide base line stereo image pairs. The algorithm has been tested extensively yielding good results over a wide range of scenes and viewpoints. A small subset of these results are presented in the paper.

• Type: Conference
• Year: 2004
• Where: ICPR
• Authors: Alexander Kaplan, Ehud Rivlin, Ilan Shimshoni
• Keywords: Computer Vision, Feature Matching Algorithm, Gray-scale Corner Detector, ICPR 2004, Significant Color Regions