Three-Dimensional Model Retrieval Using Dynamic Multi-Descriptor Fusion

作     者：Jau-Ling Shih Chang-Hsing Lee Yao-Wen Hou Po-Ting Yen 

作者机构：Department of Computer Science and information EngineeringChung Hua UniversityHsinchu 300 Pegatron Corporation.Taipei 112 

出 版 物：《Journal of Electronic Science and Technology》 (电子科技学刊（英文版）)

年 卷 期：2017年第15卷第2期

页      面：169-177页

核心收录：

学科分类：08[工学] 080203[工学-机械设计及理论] 0802[工学-机械工程] 

基　　金：supported in part by“MOST”under Grants No.102-2632-E-216-001-MY3 and No.104-2221-E-216-010-MY2 

主　　题：Index Terms--Three-dimensional (3D) model retrieval automatic relevant/irrelevant models selection (ARMS) feature re-weighting (FRW) query point movement (QPM). 

摘      要：In this paper, we propose a dynamic multi-descriptor fusion （DMDF） approach to improving the retrieval accuracy of 3-dimensional （3D） model retrieval systems. First, an independent retrieval list is generated by using each individual descriptor. Second, we propose an automatic relevant/irrelevant models selection （ARMS） approach to selecting the relevant and irrelevant 3D models automatically without any user interaction. A weighted distance, in which the weight associated with each individual descriptor is learnt by using the selected relevant and irrelevant models, is used to measure the similarity between two 3D models. Furthermore, a descriptor-dependent adaptive query point movement （AQPM） approach is employed to update every feature vector. This set of new feature vectors is used to index 3D models in the next search process. Four 3D model databases are used to compare the retrieval accuracy of our proposed DMDF approach with several descriptors as well as some well-known information fusion methods. Experimental results have shown that our proposed DMDF approach provides a promising retrieval result and always yields the best retrieval accuracy.