Research On Content-based 3D Model Retrieval Technique

With the wide application and rapid distribution of 3D models, 3D retrieval is becoming more and more necessary. Content-based 3D retrieval is an emerging research field in recent years. Retrieval techniques have been put forward and several 3D retrieval systems have been developed by different researchers. Content-based 3D retrieval technique can be classed into three types: 1) shape-based retrieval technique; 2) topology-based retrieval technique; 3) image matching-based retrieval technique. In this thesis, a detailed survey on 3D model retrieval techniques is given firstly, witch include feature extraction, similarity calculation, and database retrieval algorithms. The ways of how to research similar 3D model used in current 3D model retrieval systems are also given and a classic 3D model retrieval system is given too.Through studying other techniques, three new 3D model search methods are put forward in this thesis. These three methods all belong to shape-based retrieval technique. Before extracting feature from 3D model, we need to do some pro-processing. There are three pro-processing: Fractionize algorithm; average sampling point algorithm; normalization algorithm.There are two features used in first method: area distribution of 3D model surface and latitude orientation average radius distribution. Quadratic form distance function has been applied to calculate the similarity of two feature vectors and two passes are needed to search database in first method. Second method only use longitude-latitude orientation average radius distribution feature, L1 distance function and indirect searching database. In third method, 3D model is equally divided into 16 parts in Z axis orientation firstly. Then there are also two features to be extracted: the average distance of all points of every part and distance histogram distribution of all points in every part. The similarity calculation is same as that of second method, while the searching database is same as first method in third method. In our experiment, different 3D models can be well separated through the feature vectors extracted by these three methods. These 3D modes with similar shape have similar feature vectors.3DMS system is a Web-based 3D retrieval system, which realizes the three methods provided in this thesis. The system's search mode is to finds similar 3D models through example model, witch is simple and easy to use. The experiment result proves these three retrieval methods have a good retrieval result and fast retrieval speed, which is low 0.4 second.There are many deficiencies in current content-based 3D model retrieval techniques. Therefore, some key issues of 3D model retrieval for future research are also discussed at the end of this thesis.