Research On Technology Of Feature Extraction In 3D Model Retrieval

Media information retrieval is a heated spot in domain of computer science at present. As the fourth media, following the audio, image and video, 3D Models have been utilized in tremendous areas popularly. Research on 3D model retrieval is to be unfolding, which is attracted and valued by more and more researchers.The technology of feature extraction is the key step during 3D model retrieval process, which includes describing the 3D model as precise as possible, comparing the features in a low dimensional space as easy as possible, and getting the ideal result and destination model at last. Shape is the direct and intuitive concept to depict objects of world in human perception system. The paper focuses on the features of geometric shape on surface of 3D models. And do research on the algorithms of feature extraction.On one hand, implement the method of computing curvature in differential geometry, compute the discrete mean curvature of vertexes on surfaces of 3D mesh model, and choose the distinct vertexes by the curvature sorted. Vertexes chosen can be counted as N points, combined with the position information of them, are the new features we proposed. In the paper, we indicate the feature as( x , y , z , H) , which can be regarded as the N dimensional feature vector. Take into account the EMD (Earth Mover's Distance) method to compare the features proposed above. Introduce this similarity measure way in our experiment to verify the feasibility of our new method. For further proof, we compared the new feature with the existing algorithms, which are Shape Distribution and EGI (Extend Gaussian image) respectively, and got a conclusion that our method is efficient and has higher retrieval efficiency, epically for those models with curve pieces or extreme parts or distinct branches.On the other hand, on the basis of the global 3D model retrieval, the paper proposes a frame of 3D model partial retrieval. Through analyzing and studying the existing algorithms of 3D model partial feature extraction, the paper tried to utilize one kind of distinct feature extraction algorithm to finish the 3D model partial retrieval. There are a lot of methods about 3D model decomposition referred in partial feature extraction of 3D models. Therefore, the paper supposes to adopt the fundamental thought of the decomposition algorithm to extract the partial distinct features of 3D models and then construct the complete 3D model partial retrieval system.