3D Model Retrieval Method Based On Meaningful Segmentation

Recently,3D model has wide application in many fields. With the development of 3D modeling technology, digital scanning technology and computer network technology, there is a rapid growth in the number of sharing and reusable 3D models. So, how to manage and retrieve effectively the existing 3D models and how to improve the sharing and reuse of the 3D model resource need to be solved immediately. Based on it, the Content-Based 3D Model Retrieval (CBMR) generated, which gradually becomes a hot research field in computer graphics field.At present, the research priorities of CBMR focus mainly on its feature extraction algorithm. Existing feature extraction algorithms can be divided into four categories:(1) feature extraction algorithm based on the statistical characteristics, (2) feature extraction algorithm based on geometric transformation, (3) feature extraction algorithm based on two-dimensional (2D) image maps, (4) feature extraction algorithm based on topology structure. In describing and extracting the model shape characteristics, these algorithms mainly consider the overall shape and topology structure. They can effectively extract the overall shape features, but ignore some significant local minutiae of the models. So for the complex models or some models with similar external contour but different details, these algorithms lack of recognition and will affect the retrieval accuracy of the 3D models.In view of the problem that current 3D model retrieval algorithms describe the local details of the models inadequately, based on the minimum rules and human visual perception characteristics proposed by the psychology and psychophysics theory, this paper proposes a new 3D model retrieval method based on meaningful segmentation theory and the multi-features combined to compute model similarity comprehensively. The algorithm can be described in follows:Firstly, we use the improved watershed algorithm proposed by this paper to do the 3D model segmentation and get the collection of meaningful model components and their adjacent relationship. Secondly, describe the overall shape distribution features of the model called global features and local details features called topology tree features. Thirdly, calculate the degree of similarity between different models by the similarity calculation method based on EMD distance and the similarity calculation method based on topology tree matching. Finally, calculate the total similarity degree between the models by the way of similar degrees of weighted summation. So, we can complete the process from the whole model similarity degree measures to local similarity match, and then realize the retrieval of 3D model.Finally, we conduct experiments in the PSB model library of Princeton University to implement the retrieval algorithms proposed in this paper. The experimental results show the proposed algorithm is lowly affected by the model noise and connectivity. Moreover, the algorithm has improved the retrieval precision and the required retrieval time is reasonable. In a word, the algorithm can get better retrieval results.