Non-rigid 3D Model Retrieval Research Based On The Geometric Features

Since three-dimensional model is applied to several fields, the growth of its quantity is exponential. Searching the relative model in the large three-dimensional model database is significant for three-dimensional model designers to improve work efficiency and reduce the cost of research and development. With the development of computer graphics and three-dimensional animation, there are several types of three-dimensional model, which could be divided into rigid model and non-rigid model. In this society,non-rigid model is more appropriate for the development and application of three-dimensional model than rigid model, due to it is easy to change its shape and structure. However, non-rigid model retrieval is more and more difficult, owing to its polytrope.Since the geometrical characteristic of geodesic distance have equidistant invariance,it is efficient to cope with non-rigid model to apply to retrieval. However, sampling point of algorithm on model surface is not satisfactory, and it cannot contain a larger number of model shape information and equilibrium distribution. It means that it is not sufficient to describe the model shape by using model features on the sampling points, which will cause decrease the accuracy of retrieval. What’ more, anomalous points will disturb the matching between models, and it will have not good influence on model retrieval. Hence,this paper will research the problems as mentioned before and propose an interest points equilibrium selection algorithm based on the shape feature. Comprehensive three shape characteristics to calculate vertices’ characteristic parameters, and by means of the way in equilibrium selection to select interest points that contain a large number information of model shape and are equilibrium distribution. Moreover, this paper also propose a Hungarian matching algorithm of eliminate anomalous points. This algorithm will help to deal with the noise data, eliminate the interference of anomalous points, and obtain a more accurate optimal matching relation about interest points and similarity distance between two models. What’s more, this paper will improve the bag of geodesic histograms feature extraction algorithm, solve the problem of data redundancy due to a large number of statistics data of geodesic distance, and provide a better model feature descriptor.As mentioned before, this paper designs and realizes a non-rigid three-dimensional model retrieval system based on geometric features which could effective retrieval the similar non-rigid models and the retrieval precision is also improved.