Accurate Disparity Estimation Based on Integrated Cost Initialization

Disparity estimation is one of the most important research topics in computer vision. Numerous local-based approaches have been proposed to solve this problem. Among them, most state-of-the-art methods mainly focus on color information when initializing the cost volume. However, color signals are less robust and more easily affected by image noise, illumination variation and radiometric differences. In this paper, we develop a high quality disparity estimation system based on an integrated matching cost initialization algorithm. During the cost initialization step, three individual cost terms are utilized to construct the cost volume: Gradient-based Census Transform (GCT), Absolute Color Differences (ACD), and Gabor Pattern Differences (GPD). The proposed method produces impressive performance and ranks excellently in the Middlebury benchmark. Furthermore, we present that the proposed scheme is also capable for real-world outdoor scenes which contain many challenges. Both quantitative and qualitative evaluations demonstrate that our approach is currently one of the most accurate local-based stereo matching algorithms