Shape Similarity Assessment Based On Partial Feature Aggregation And Ranking Lists

Zhenzhong Kuang,Zongmin Li,Yujie Liu,Changqing Zou

DOI: https://doi.org/10.1016/j.patrec.2016.05.026

IF: 4.757

2016-01-01

Pattern Recognition Letters

Abstract:In this paper, we focus on the problem of similarity assessment of isometric 3D shapes, which is of great relevance in improving the effectiveness of retrieval tasks. We first present an effective shape representation technique by proposing a partial aggregation model based on the bag-of-words paradigm. This technique can effectively encode our multiscale local features and has a good discriminatory ability. We then develop a parameter-free distance mapping approach to re-evaluate the similarity results based on intrinsic analysis of a well organized reciprocal k-nearest neighborhood graph. Different from the existing methods which determine k manually and globally, the proposed method can automatically adjust k to a reliable local domain, which therefore ensures a more accurate similarity measurement. We fully study our shape representation technique and evaluate the performance of the proposed distance mapping approach on several popular public shape benchmarks. Experiment results have demonstrated the state-of-the-art performance of our approach. (C) 2016 Elsevier B.V. Allrightsreserved.