Human 3D Model Retrieval Based On Skeleton Structure

3D model is the fourth digital media technology after sound,image and video.With the rapid development of multimedia acquisition equipment and 3D modeling technologies,the quantity of 3D model is growing explosively.In the face of vast amounts of 3D models,how to manage them effectively,and realize the efficient and convenient 3D model retrieval and information acquisition becomes the urgent problem to be solved.3D model skeleton can not only represent the shape information and topology skeleton structure information of the model,but also can be regarded as a compression technique of the original model,which can greatly reduce the storage space and improve the utilization of computer storage space.With the development of graphics and visualization technology,3D skeleton has become an important research direction in the 3D model retrieval.At present,the 3D model skeleton has been widely used in computer aided mapping,human organ reconstruction,virtual reality,animation simulation,industrial manufacturing,digital entertainment and other fields.Based on the skeleton extraction algorithm of 3D model,this paper presents an innovative 3D model retrieval algorithm.The main contents and innovations of this paper are as follows:1)Database Construction: In the existing 3D model database,some complex 3D models from some existing 3D model database are selected and a new database is built.2)Skeleton Extraction: In this paper a L₁-medial skeleton extraction algorithm is proposed.And several Optimization algorithm such as: Sampling Optimization via topologic subdivision,Distribution optimization via density-based weighting,and Position optimization via ellipse fitting are adapted as for some defects limitations of the original models.3)Matching Algorithm: A Hyper-graph matching algorithm based on reweighted random walking is proposed.This method adopts high-order tensor and random walking to Hyper-graph matching.Different order tensors express the different order hyper-edges' information,and formulate the lower-order hyper-edges into a signal high-order tensor in a recursive manner.It's an effective constraint on hyper-graph structure.The proposed algorithm achieves a state-of-the-art performance showing robustness to both deformation and outlier noises.Finally,comparative experiments on the new database show that the proposed method clearly outperforms existing matching algorithms.