Research And Implementation Of Feature Extraction And Relevant Feedback Of 3D Model

The rapid development of the three-dimensional (3D) modeling technology and Internet technologies has led to an increasing amount of 3D models, both on the Internet and in domain specific database. And 3D models play a very important role in many fields, such as molecular biology, cultural heritage protection, computer-aided design and manufacturing and so on. Thus, this paper arrive accuracy and efficiently retrieve for the mass of 3D models, through researching the technology of feature extraction and relevance feedback. The main work and progress are listed as follows:1) Implement the improved PCA and isotropic preprocessing algorithm, and we also compare and analyze the experimental results. The experimental results show that preprocessing 3D models before the feature extraction effectively improves the accuracy of the search algorithm.2) Improve an extraction algorithm based on Terminal Projection Transformation. From the experimental results, compare with a simple extraction algorithm of the pixel distribute histogram based on three views projection, the algorithm has a good ability to distinguish different 3D models, and effectively improves the accuracy of 3D model retrieval.3) Propose a new feature extraction algorithm for 3D model based on Range Image. First, this algorithm computes 3D models projection range images on six directions, and then computes edge direction histogram and Zernike moment of these range images. At last combining the two feature matches 3D models. The experiments show that the algorithm avoids the traditional visual shortcomings of the loss of spatial information, and improves the accuracy of the retrieval and robustness effectively.4) This paper introduces the weight SVM to relevance feedback because of the semantic gap between the high-level semantics and low-level features. The experimental results indicate that the algorithm eliminates the semantic gap using manual labeling, and is good at dealing with the number unbalanced problem of "relevant" and "irrelevant" training samples, and improves the results of 3D model retrieval effectively.5) This paper designs and develops a 3D model retrieval system with human-computer interaction function, which is an experimental system. This system have completed 3D retrieval task well.This paper is supported by National High Technology Research and Development Program(863 Program. No.2008AA01Z301) of China.