3d Model Retrieval Based On Local Shape Distribution Algorithm

The number of 3d models in the electronic world is exploding with the development of techniques for 3d scanning and 3d modeling. Search engine is a necessity to get useful information from so many 3d models. While text-based search systems have some limitations in describing and searching 3d models, content-based systems offer an effective solution.There are many content-based 3d model search algorithms now. We divide them into four categories, which are global transform-based, histogram-based,2d view-based and graph-based. Most of the algorithms at present need pose normalization and few of them have the ability of local description of the models.We present a new 3d model retrieval algorithm in this paper, the local shape distribution-based algorithm.The local shape distribution method uses the distribution of local geometric features as local description of the 3d model. For a vertex P on the mesh of the 3d model, we sample some points randomly around P at first. Then, calculate shape function values for these points. The distribution of the shape function values is the local shape distribution. In this paper, we choose the AD (angle-distance) shape function, of which the local shape distribution can be described as an image that we called PODI (Point Description Image).We use the local shape distribution to construct the 3d model descriptor PODISet for model retrieval. First, choose several key attributes, and then calculate the key attributes values for the vertices on the mesh of the 3d model. For each key attribute, the vertices with the largest Nr and the smallest Nr values are chosen as key vertices, where Nr is a parameter of the algorithm. At last, calculate PODI for the chosen key vertices, and we get a PODI set, called PODISet.As a 3d model descriptor, PODISet does not need pose normalization, and it has invariance, efficiency and robustness. Besides, it has the ability of local description.We select proper parameters for the proposed method in the experiments on PSB. And we compare the retrieval performance of PODISet with other methods. The experiment shows that PODISet has better performance.