Abstract: Aiming at the problem that the similarity measure of most stereo matching algorithm was mostly based on gray characteristics of pixels, which could not completely eliminate the differences of matching, and was prone to ambiguity, we proposed an improved matching algorithm based on nonparametric Census transform. It increased the measure of local texture contrast value in the traditional nonparametric Census matching process, and the directionality of the image texture was introduced, so that gray value of the center pixel was no longer the sole determinant, and matching template of the transform mode was improved, thus the ambiguity problem of traditional matching algorithm was effectively solved. The experimental results show that the improved algorithm is an effective and reasonable method of stereo matching, which improves the accuracy of dense matching.

Key words: visual detection, stereo matching, nonparametric Census , transform, matching accuracy