科研之友
微信
新浪微博
Facebook
分享链接
摘要
Stereo matching is the key of reconstructing 3D face model from 2D face images and Barnard's algorithm is one of the classical stereo matching algorithms based on feature points. In order to enhance the stability of extracting feature points, reduce computational complexity and minimize the effect of subjective factors, in this paper, Barnard's algorithm of edge point matching was improved from two aspects: edge point sets of left and right face images were extracted with dyadic wavelet transform, rather than comparing gray values of adjacent pixels; on the other hand, edge points in binocular stereo face images were matched under epipolar constraint in parallel binocular stereo matching. Experimental results demonstrate that stability and efficiency of stereo matching algorithm are both improved.
