Research On 3D Model Retrieval Based On Global Feature And Local Feature

In recent years, with the rapid development of three-dimensional animation and Internet technology,3D model becomes a new type of multimedia data. The main task of the 3D model retrieval is to find the model that the users need from large numbers of 3D models quickly and effectively. Content-based 3D model retrieval technology has received widespread attention, hence becoming an active field in research.In this paper, with focus on the analysis of the feature extraction which is the key technology of 3D model retrieval, researches and analyses of global feature extraction algorithm based on D2 shape distribution are made firstly. Aiming at the problem that this method can describe the whole shapes and characteristics of the model while ignore the local characteristics'information, this paper presents two local feature extraction algorithms. One is sub-block shape distribution based on concentric spherical shell segmentation and the other is distance distribution statistical vector. With this method, sub-block is obtained by means of concentric spherical shell segmentation, then the feature of sub-block is extracted, and the global similarity is obtained by summing of weighted sub-block similarity, and finally realize 3D model retrieval. This method enhances the importance of local features to the model retrieval.The experimental results also show that the local matching method has a higher precision under the same recall rate, and improves the retrieval performance.In addition, this paper presents an optimization retrieval method that combined with global feature and local feature. In this method,first the global feature extraction algorithm is used to retrieve the model that is similar to the whole of the query model from the model databases,then the local different models are filter out, so as to retrieve the model which has the high correlation both in whole and local parts. The experimental results show that the proposed method is more efficient than the global feature extraction algorithm and the local feature extraction algorithm as well as their fusion algorithm, at the same time, achieving higher accuracy under the same recall rate.