A Multi-Decision Area Based Approach for Solving Correspondence Problem in Stereo Images with High Accuracy

In computer vision and digital photogrammetry solving the correspondence problem is one of the most important field of research. Conventional methods used to generate dense density maps for stereo image pairs is classified as either area based or feature based methods. In this paper a multi decision area based approach has been proposed that performs pixel matching in stereo-pairs. The approach makes use of various matching methods at different stages depending upon the value of parameter that has been used as Decision Factor. Normalized Cross Correlation which is a measure of similarity between windows has been used as decision factor, if maximum Normalized Cross Correlation calculated exceeds a certain upper threshold it is considered as If Normalized Cross Correlation value lies in midrange, 2 nd stage analysis is performed that makes use of Edge Analysis and Euclidian distance while spline fitting is employed for lower values of decision factor, if NCC obtained falls below lower limit, no match is obtained. With the multistage approach & parallel processing speed of dense disparity map generation will improve along with improved accuracy. The method is tolerant to geometric distortions. The percentage of accurately matched pixels in stereo pairs was found to be 94.15% in one of the test data set, which was very high compared to standard Normalized Cross Correlation based approach (72%).