A Robust SIFT-based Algorithm for Stereo Matching by Combining Region-based and Feature-based Methods

Abstract

Recently, many researchers are actively studying on the reconstruction of 3-dimensional structures from the 2- dimensional images acquired from a stereo camera. However, the disparity map calculated from the existing methods does not always provide accurate depth information by the false matching and occluded areas. In this paper, by simultaneously applying the feature-based methods [1] and region-based methods[2], we effectively solve the problems caused by the false matching and occluded regions[3]. By experimental results we show that the proposed method can generate more accurate disparity map even in the case of occluded area or complex texture area [4].