Underwater stereo-matching algorithm based on belief propagation

Yongbing Xu, Dabing Yu, Yunpeng Ma, Qingwu Li, Yaqin Zhou

Abstract: Using stereo-imaging systems to collect 3D information is innovative and flexible for underwater exploration. The stereo-matching of underwater image pairs is a significant and challenging task due to the poor visibility and the complex underwater light field. In this paper, we propose a novel underwater stereo-matching algorithm based on belief propagation(BP). We design the energy function suitable to apply in the underwater scenes. Specifically, we use zero-based normalized cross-correlation and Hamming distance to form the data term that computes a measure of similarity between points of the binocular image pair and design the smoothness term based on the color metric to settle the discontinuity of the disparity map. Furthermore, we use bilateral filtering to gather the initial matching cost and propose a filling operation for the occlusion in the disparity map. Extensive experiments demonstrate the effectiveness of the proposed algorithm both on simulated UW-Middlebury dataset and real-world underwater images pairs