Research On 3D Model Retrieval System Based On Multi-feature And Relevance Feedback

In recent years, people have created mass 3D models and huge 3D model databases have appeard. People always desire to work on similar model rather than create a new model out of thin air. So 3D model retrieval engine, helps people find similar model. is becoming more and more necessary. How to build an efficient 3D model search engine become urgent.In 3D model retrieval and recognition research, 3D shape analysis and matching is the first and fundamental task. Mostly, people use similarity between 3D models by analyzing their shapes to distinguish similar models.There have been dozens of shape features and we select several features to study: shape distribution, thickness histogram, spherical harmony. After studying shape distribution, a method differentiates 3D models using point-pairs, we find that engine can separate the 3d models well through classification of point-pairs.We present a shape-based 3D search engine using relevance feedback algorithm. In the recent years, someone used multi-feature to search 3D models by combining the features with weights. But it's difficult to make sure the weights on the feature vector. In our system, a relevance feedback algorithm is used to automatically determine the most appropriate weights according to the user's response. Furthermore, we create semantic correlation matrix, which is made of correlation strength. By using the matrix, it's easily to know the correlation between models. Intuitively, the larger the correlation strength is. the more likely that these two models are semantically similar to each other.This paper designs a 3D retrieval system—M-3D using multi-feature and retrieval algorithms. This system have completed 3D retrieval task well. Moreover, some key issues for future research are also discussed at the end of this thesis.