Multishape-features and text-feature integration on 3D model similarity retrieval. (21st July 2008)

• Record Type: Journal Article
• Authors: Akbar, Saiful; Kung, Josef; Wagner, Roland
• Is Part Of: International journal of innovative computing and applications. Volume 1:Number 3(2008)
• Page Start: 171
• Page End: 184
• Publication Date: 2008-07-21
• Subjects: 3D model retrieval -- feature integration -- distance integration -- shape features -- text features -- text-feature integration -- similarity retrieval -- shape-feature vector; Evolutionary computation -- Periodicals; Neural networks (Computer science) -- Periodicals; Genetic programming (Computer science) -- Periodicals; Biologically-inspired computing -- Periodicals; Swarm intelligence -- Periodicals; Quantum computers -- Periodicals; 006.3
• Languages: English
• ISSNs: 1751-648X

Abstract

In this paper, we describe several 3D shape descriptors and integrate a textual descriptor with them for 3D model retrieval. We analyse five Shape-Feature Vector (FV) integration approaches, namely Pure FV Integration (PFI), Reduced FV Integration (RFI), Weight-Associated RFI (WRFI), Distance Integration (DI) and Rank Integration (RI). By running all possible-combinations of weighting factors on a training data set, the best weighting factor for each approach is obtained. Our experiments show that the best weighting factors improve the retrieval performance on not only the training data set, but also other data sets. This paper also shows that Distance Integration delivers the best retrieval effectiveness and Reduced FV Integration has the capability to deal with unknown query. In addition, the Distance Integration provides faster processing as it uses precomputed pair-wise distance and is more advantageous than PFI because of the dimension reduction. Hence, the use of both approaches (DI and RFI) is proposed. This paper also explains a use of the model file name as the only resource for text feature extraction. We study several textual similarity measures and then integrate multishape and text features into 3D model retrieval. Our experiments show that text feature can discriminate 3D models to each other in a certain degree of effectiveness, and integration text feature with multishape-feature improves retrieval effectiveness.