Research On 3D Model Retrieval Technology Based On Multiple Feature Fusion And 2D Views

3D models have been used widely in many fields,and the demand for them is also increasing.As a result,the number of 3D models has growing rapidly.It takes a lot time and effort to create a 3D model with high fidelity.Therefore,how to retrieve the same or similar 3D models from existing 3D model databases quickly and accurately has become a hot pot in recent years.The major and difficulty of 3D model retrieval technology is feature extraction.As a single feature can not fully describe a 3D model,more and more researchers paid attention to multiple feature fusion.3D model retrieval methods based on multiple feature fusion can describe a 3D model more accurately and improve the accuracy of retrieval results.However,how to improve the description ability of existing methods and propose more effective retrieval methods is a problem.In addition,how to select features for fusion is a difficulty worth studying.For these problems,our research can be summarized as following two aspects:(1)We propose a novel 3D model retrieval method fusing global and local information.Several 2D projected views are obtained from different directions by rotating,next we extract edge feature for global information by Canny algorithm and word frequency vector feature for local information by bag-of-feature based on SIFT features respectively,then these two features are fused into a new feature for 3D model.The proposed method focuses on the global and local information,which can improve the retrieval accuracy well.(2)We propose a 3D model retrieval method based on manifold ranking,global and local information.Most existing methods use features of 2D projected views to represent a 3D model directly but ignore their contributions to a 3D model,which reduces the accuracy of a 3D model feature and affects the retrieval results.Manifold ranking can solve this problem.Therefore,we propose a 3D model retrieval method based on manifold ranking,global and local information.Several 2D projected views are obtained from different directions by rotating,next we extract edge feature for global information by Canny algorithm and word frequency vector feature for local information by bag-of-feature based on SIFT features respectively,then we compute the weights of 2D projected images by manifold ranking algorithm,finally these two features are fused into a new feature for 3D model.The proposed method focuses on contributions of 2D features,the global and local information,which can improve the retrieval accuracy well.We do a series of experiments on PSB and SHREC2012 GTB for these two proposed methods,and the experimental results verify the validity and advantages of them.