Improved stereo correlation using Moravec operator and Kolmogorov-Smirnov test

Extraction of effective depth information from a stereo pair is a classical image processing problem. In the present paper, the performance of a mean intensity level invariance based template matching algorithm is improved using Moravec operator and Kolmogorov-Smirnov test. In the original template based stereo-correlation algorithm, the entire images are compared using correlation and search windows for the purpose of extracting useful depth information. However, it is generally observed that the entire stereo pair is not a good source of depth information. For the purpose of identifying potential areas containing depth information like the edges and corners, Moravec operator is employed followed by an iterative thresholding procedure. On the other hand, Kolmogorov-Smirnov test is applied to extract areas of comparable intensities. The incorporation of this statistical test thus reduces false comparisons between the two images forming a stereo pair at the time of identifying depth information. Furthermore, since the Kolmogorov-Smirnov test is a nonparametric hypothesis testing method, in order to apply the test, we don't need to assume any specific intensity distribution for the stereo pair. The performance of the stereo correlation algorithm is found to be improved with the incorporation of the above two tools/techniques. Experimental results on real stereo pairs are presented.