3D Model Retrieval System Based On Multi-feature Fusion

With the widely using of the multimedia in the internet, the resources of multimediainformation are gradually increasing. How can retrieve the relevant model from massivelibrary of3D model have a strong practical significance to help the3D model designer toimprove worker productivity and reduce development costs. The most crucial task of the3D model retrieval is to extract the efficient and comprehensive feature descriptor. But now,there is not a universal feature extraction algorithm can achieve good retrieval results on alltypes of hybrid models. Analysis of its causes, because the3D model itself contains amodel of the shape of the distribution in3D space, the complex branching structure andgeometric topology information, etc. Single feature descriptor only extracts some of thecharacteristics from a certain point of view3D model with some limitations. Theexpression is not100%comprehensive3D model itself, thereby reducing the retrievalperformance.Therefore, based on the study and the realization of the distribution shape featureextraction algorithm, extraction segmentation algorithm based on statistical characteristicsof the sphere as well as the latest3D model retrieval outstanding performance Surface Areadescriptors on global competition SHREC2014, this paper focus on research formulti-feature fusion is difficult to determine the weights of difficulty, drawing on entropyto determine the field of application experience in the evaluation of multi-index weights,proposes a multi-feature information based on entropy weights determined fusion program.Experimental results show that recall, the precision and R-Precision and other indicators ofthe fusion based on information entropy search results are better than single featureextraction algorithm, ideal retrieval results.In this paper, I design and implement a3D model retrieval system based multi-featurefusion, The system uses Structs2.0, WebGL, html5and other web developmenttechnologies, based on B/S architecture, it has feature extraction and matching with asingle retrieval (shape distribution, Surface Area, statistical characteristics of the sphere),multi-feature fusion search, the search results display3D rendering, and other functions.