Non-rigid point set registration using dual-feature finite mixture model and global-local structural preservation. (August 2018)

• Record Type: Journal Article
• Authors: Zhang, Su; Yang, Kun; Yang, Yang; Luo, Yi; Wei, Ziquan
• Is Part Of: Pattern recognition. Volume 80(2018:Aug.)
• Page Start: 183
• Page End: 195
• Publication Date: 2018-08
• Subjects: Registration -- Dual-feature -- Finite mixture model -- Regularization -- Structural preservation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.03.004
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We develop a dual-feature finite mixture model based point set registration method. We define a local structure descriptor. We constrain the energy function using both the global and local structural regularization terms, preserving the structure at both global and local scales during the registration. The performance of our method outperforms state-of-art methods in this area. Abstract: We present a dual-feature based point set registration method with global-local structural preservation. A finite mixture model which is able to deal with two features is first constructed. The mixture structure descriptor (MSD) is then obtained by smoothly combining the local structure descriptor (LSD) with the original coordinates through an annealing scheme. Substituting the MSD into the constructed model a fuzzy corresponding matrix is acquired. Next, the energy function which contains three main terms is formulated in the reproducing kernel Hilbert space. The first term is based on the L 2 estimation ( L 2 E ) criterion, and the other two terms play a complementary role to improve the robustness and accuracy for transformation estimation at both global and local scales. The performances of our method in synthetic data, sequence data and real data against nine state-of-the-art methods are tested where our method shows favorable performance when compared with other methods.