Stereo analysis of low textured regions with application towards sea-ice reconstruction

Images with large areas of low texture pose significant challenge to stereo algorithms. We propose a novel segmentation based stereo scheme tuned to handle such scenes. We combine entropy based filtering and two levels of belief propagation to create an algorithm that can handle a scene with very little color variation. We test the performance of our method on standard stereo evaluation and find that it outperforms comparable algorithms. We also present results on dataset containing images of a frozen river and find that the algorithm produces good disparity estimates.

• Type: Conference
• Year: 2009
• Where: IPVC
• Authors: Rohith MV, Gowri Somanath, Chandra Kambhamettu, Cathleen Geiger
• Keywords: Belief Propagation, Dense Disparity, Entropy Filtering, IPVC 2009, Sea Ice Stereo, Segmentation Stereo, Textureless Stereo