Color Invariant Census Transform for Stereo Matching Algorithm

Illumination and color invariance are important problems in computer vision (CV). The Census transform (CT) can resist change of illumination intensity and is widely used for many applications in CV and consumer electronics. However, only grayscale images can be processed in the CT algorithm. In this paper, a new color Census transform (CCT) based on a color invariance model for stereo matching is proposed because color images are with more significant features and most source images are color. To evaluate the proposed method, its resulting disparity maps and computation time are compared with grayscale modified CT (MCT). Experimental results show that the computation time of the proposed method is little more than the grayscale MCT. But the proposed method is able to significantly improve structure features of disparity maps, compared to the grayscale MCT. Further, the CCT is not to be affected by shadows, highlights, and variations in illumination.