Image Matching Using Mutual k-Nearest Neighbor Graph

• Published in: Intelligent Computation in Big Data Era pp. 276 - 283
• Main Authors: Li, Ting-ting, Jiang, Bo, Tu, Zheng-zheng, Luo, Bin, Tang, Jin
• Format: Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2015
• Series: Communications in Computer and Information Science
• Subjects: Image matching; mutual k-nn graph; spectral technique
• ISBN: 3662462478; 9783662462478
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-662-46248-5_34

Though weighted voting matching is one of most successful image matching methods, each candidate correspondence receives voting score from all other candidates, which can not apparently distinguish correct matches and incorrect matches using voting scores. In this paper, a new image matching method based on mutual k-nearest neighbor (k-nn) graph is proposed. Firstly, the mutual k-nn graph is constructed according to similarity between candidate correspondences. Then, each candidate only receives voting score from its mutual k nearest neighbors. Finally, based on voting scores, the matching correspondences are computed by a greedy ranking technique. Experimental results demonstrate the effectiveness of the proposed method.