Complex Correlation Statistic for Dense Stereoscopic Matching

A traditional solution of area-based stereo uses some kind of windowed pixel intensity correlation. This approach suffers from discretization artifacts which corrupt the correlation value. We introduce a new correlation statistic, which is completely invariant to image sampling, moreover it naturally provides a position of the correlation maximum between pixels. Hereby we can obtain sub-pixel disparity directly from sampling invariant and highly discriminable measurements without any postprocessing of the discrete disparity map. The key idea behind is to represent the image point neighbourhood in a different way, as a response to a bank of Gabor filters. The images are convolved with the filter bank and the complex correlation statistic (CCS) is evaluated from the responses without iterations. The magnitude of CCS measures the image similarity and the phase gives the sub-pixel position. Our experiments show that CCS has even better sampling invariance and discriminability properties than the popular Birchfield-Tomasi dissimilarity, and the sub-pixel accuracy is higher than the Maximum Likelihood sub-pixel disparity window fitting disparity correction method we compared with. Results are shown on both synthetic and real data.