Research On Algorithm Of 3D Model Retrieval Based On Shape Histogram

With the improvement of computer hardwares, especially, 3D graphic hardwares and scanning devices. 3D models from a wide range of sources are demanded. In the content of internet, the key to 3D model retrieval is to get invariance of geometric transformations of rigid bodies. 3D model retrieval is the new development of the content-based 2D image retrieval. The main problem of 3D model retrieval is the low search accuracy rate. This dissertation focuses on the shape histogram extracting method of 3D models and its applications in the 3D model retrieval.Firstly,the advantages and disadvantages of the Principal Component Analysis(PCA) was analysised. The uncertainty of the principal axes and their orientation produced by 3D Model Retrieval PCA preprocessing is the main factor affecting the retrieval result. A novel pretreatment method using the triangle center of the mass weighted principal component analysis was proposed to complete the pose normalization.Secondly , the feature extraction method proposed by Ankerst was researched and analysised. As there was no topology information and important shape features describing the model inadequately, the results of retrieval is not ideal. A novel feature extraction based on Ankerst's shape histogram method was proposed. Topological features were fused into it. Model features can be descripted sufficiently.Thirdly,similarity measure functions were constituted for our model features. The similarity of shape histogram was calculated by using the quadratic function of distances. Then the similarity of topological features was computed by using overlapping projection method. Weighted sum of the two parts was determinated as the result. Finally,an experimental system was designed to test the improved algorithm. The improved precision was compared to Ankerst's algorithm. There are many deficiencies in current content-based 3D model retrieval techniques. Therefore, some key issues of 3D model retrieval for future researches are also discussed at the end of this thesis.