Census Filtering Based Stereomatching under Varying Radiometric Conditions

Census transform is a non-parametric local transform which is based on the relative ordering of pixels instead of the pixel intensity values within a window; therefore it is insensitive to radiometric variations like exposure changes and illumination variations. This work focuses on an efficient census filtering based local steromatching for images taken under varying radiometric conditions. For matching cost value computation hamming distance is taken for the dissimilarity measure. Finally for cost aggregation guided filtering is used. Guided filter has edge preserving property and runs faster than the bilateral filtering based cost aggregation. The experimental results shows that the proposed algorithm provides better disparities for images taken under varying radiometric conditions than the conventional stereomatching algorithm.