Robust Stereo Matching Combining SIFT Descriptor with NCC under MRF Framework

There have been several stereo matching methods that perform well under the circumstance of color consistency. However, various factors including radiometric and device variations between images will drop color consistency and then seriously degrade the performance of those methods. In this paper, we propose a robust method to cover these situations. We use a measurement combining SIFT descriptor in intensity space after color histogram equalization with Normalized Cross Correlation (NCC) in color invariant log-chromaticity intensity space to compute the cost of point correspondence. The measurement ensures that illumination-independent gradient information and color invariant correlated information are integrated properly. We evaluate our method and find it outperform state-of-the-art algorithms, in particular on the datasets for radiometric variations.