Measurement of stereo matching on images using dissimilarity estimation

This paper proposes the dissimilarity based adaptive prior modelling for stereo images and improve the performance on images. The fast likelihood function based on rank transform implemented in matching images that replaces the intensity of a pixel with its rank among all pixels within a certain neighbourhood. Rank transform reduce the radiometric gain, bias also reduce the discriminatory power in stereo matching, by giving matching result. Adaptive prior modelling is proposed, improves the smoothness of matching result and is defined as a pixel wise energy function by using adaptive interference between neighbouring disparities. The disparity is estimated by minimizing the joined energy function which combines the likelihood matching and prior modelling. The dissimilarity based adaptive prior modelling measures the dissimilarity between joint energy function and gradient (GRAD). The optimal weight ω is determined between join energy function and gradient (GRAD) by maximizing the number of reliable correspondences that are filtered using cross checking test. Finally we consider filters like wiener, median, order static filter to improve the smoothness of stereo images.