Research Of 3D Model Retrieval Technology Based On View

With the rapid development of 3D scanning technology, 3D graphics modeling methods and computer hardware technology, the rapid growth of 3D model is not only in quantity, but its applications are becoming more widespread, such as computer aid design(CAD), virtual environment, 3D games etc. Because creating a realistic 3D model requires a lot of time and effort , and lots of 3D models can be shared in the Internet. So it is becoming an urgent issue to help users find their desirable 3D models accurately and efficiently from Internet or model databases. The 3D model retrieval technology is a kind of important method.This paper introduces relevant knowledge of 3D model retrieval, and analyzes the existing 3D model retrieval technology. The traditional method using 2D views for 3D model retrieval will cause redundant view problem obviously. This kind of method based on redundant views can acquire better retrieval accuracy, but the matching cost is very expensive. This paper proposes a 3D model retrieval algorithm based on single view and an image-based retrieval system for 3D model is implemented. The main research work in this paper are below:1. In order to solve the redundant problem of views in the existing technology. We propose an approach based on a single view for 3D model to measure shape feature. The algorithm consists of three steps as follows: Firstly, the pose of 3D model will be adjusted and the main view through rendering the 3D model that can best express the shape feature of 3D model is gotten. Secondly, outline of the main view is sampled and can be described by inner distance and inner angle. Finally, dynamic programming algorithm is used to compute similarity among different 3D models. Experimental results show that the proposed method can obtain higher accuracy rate of model retrieval.2. The existing best view measurement is difficult to apply for various different kinds of 3D models. This paper proposes a distinct algorithm for similarity learning based on samples. The best view of 3D model will be selected according to the similarity metric. In this algorithm, the best views defined by user will be taken as training set. Through AdaBoost algorithm for supervised learning, the best view selection model with user knowledge can be constructed. Experimental results show that the best view gotten from this method will effectively close the user's selection results, and has good stability and generalization ability.3. We design and implement a 3D model retrieval based on images. This system can extract the foreground object according to the user's choice, then 3D model retrieval can be implemented through foreground object. This system also has good human-machine interaction, and provides a good platform for follow-up research.