Dense image correspondence under large appearance variations

• Main Authors: LIU, Linlin, LOW, Kok-Lim, LIN, Wen-yan
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2013
• Subjects: belief propagation; image matching; image motion analysis; image registration; SIFT Flow; Graphics and Human Computer Interfaces; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/4811; https://ink.library.smu.edu.sg/context/sis_research/article/5814/viewcontent/image_match_icip2013.pdf
• Institution: Singapore Management University

This paper addresses the difficult problem of finding dense correspondence across images with large appearance variations. Our method uses multiple feature samples at each pixel to deal with the appearance variations based on our observation that pre-defined single feature sample provides poor results in nearest neighbor matching. We apply the idea in a flow-based matching framework and utilize the best feature sample for each pixel to determine the flow field. We propose a novel energy function and use dual-layer loopy belief propagation to minimize it where the correspondence, the feature scale and rotation parameters are solved simultaneously. Our method is effective and produces generally better results.