Stereo Matching Based on Local Invariant Region Identification

Abstract

Among the solutions for stereo matching which is regarded as one fundamental topic in computer vision, methods based on regions can overcome the drawbacks of methods using points and lines as the features since regions can more exactly reflect the inherent properties of images processed. In this paper, we propose an efficient method to extract the local invariant region, in which the affine moment invariant is also combined. Experiments prove our method is effective while easier to implement.