DISPARITY MAPS EXTRACTION BASED ON MRF METHOD FROM AN IMAGE SEQUENCE by Lan Jun Thesis Supervisor: Assistant Professor Liming Zhang Master of Science in Software Engineering The research topic of automatic extracting disparity/depth cues from images is great practical significance. The applications like robot navigation, 3D reconstruction and 2D to 3D video conversion always need disparity/depth cues. This paper presents a novel method based on MRF method to estimate disparity maps of image sequences. The main contributions of this dissertation are summarized as follows: (1) Make a survey about the status and research methods of disparity extraction, and focus on research and analysis of MRF based methods. (2) Unlike previous methods which usually choose whole disparity range to minimize the global energy function, our method automatically employs priors to decide disparity range for each point so that the running time of convergence of global energy function is reduced. (3) Our method uses camera model to improve the precision of cost of matching so that the distribution of restored disparities become accuracy. (4) Our method utilizes geometry coherence to recover the disparity of occluded regions in one input view The high-quality recovered disparity maps selected from challenging image sequences show the effectiveness of our method. The acceptable running time of recovered disparity maps demonstrate the performance of our method.