3D Model Retrieval Method Based On Shape Weighted

With the multimedia technology and virtual reality technology goes to improve, the three-dimensional (3D) model are widely used in many areas of medicine, mechanical engineering, computer-aided design (CAD) and entertainment. The growing popularity of the3D model application, the quantity of the3D model is exponential growth, which greatly enriched the3D model database. Describing3D model needs large amount of information, complex shape and other properties similar relationship between models, result in the rational organization of the3D model of the database is very difficult. At the same time, in order to fully utilize the existing model resources, and quickly find the3D model, there are great demands of the3D model database and3D model retrieval methods.Content-based3D model retrieval, firstly, calculate automatically from the model data and extract the features of3D model, such as shape, spatial relations, color and texture of material, to establish3D model of multi-dimensional information index. Secondly, calculate the similar degree between the query model and the target model in a multi-dimensional feature space, and then realize the3D model database browsing and retrieval. The retrieval technology bases on similarity matching to find the visual characteristics of the3D model required by the user, which is closer to the way of people use information in real life.Feature extraction and similarity measure are the key technologies of content-based3D model retrieval, also are the emphases and difficulties of the research. Around the two key technologies, this article conducts the research from three aspects:(1) Feature extraction:Many existing feature extraction algorithms pay more attention to the model of high-level information (such as topology, voxel grid, etc.), and less consideration on the model surface geometry information, which make multiply algorithm computation, and only slowly improve retrieval efficiency. According to this problem, bases on statistical-distribution feature extraction algorithm, puts forward a3D model of feature extraction algorithm based on shape weighted. The algorithm firstly calculate the distance between sample points and the center point, then take into the sample points in the triangles area percentage as a weight, to build characteristic histogram and be the model shape descriptor. Experiment shows that, the extracted shape descriptor, which has invariance in translation, scaling, and rotation, etc, the algorithm has simple steps, and low time complexity.(2) Similarity measurement: Feature histogram similarity measure is based on the feature extraction of model, and calculates the distance between the shape descriptors, which is equally important to the model retrieval. There are many characteristics matching methods (similarity measurement method), often used for distance measurement in the field of3D model retrieval method are Euclidean Distance, Manhattan Distance, Minkowski Distance, Hausdorff Distance, Histogram Intersection Distance, Quadratic Form Distance and EMD Distance. These distance measurement methods are not the same apply on the high dimension feature descriptor matching, and have deficiencies for the deformation models. According to this problem, proposes a feature histogram window translational distance algorithm. By calculating value differences between sample model features histogram and database model feature histogram’s corresponding windows, and values differences between different distance windows, get the average difference based on the distance between the windows as weighted, which will be as the similarity measure between the model standards. Experiment shows that, the method has high robustness to the model deformation and histogram deformation, has strong resistance to noise and good stability.(3)3D Model Retrieval Method Based on Shape Weighted: Respectively using my feature extraction algorithm based on shape weighted to obtain shape descriptor, and my feature histogram window translational distance algorithm to measure similarity between models, to realize3D model retrieval method based on shape weighted. Experiment shows that, the retrieval method has quick response; recall and precision are obviously improved.