Semantic-driven 3D Model Component Analysis And Shape Modeling

This paper focuses on the semantic-driven 3D model component analysis and shape modelling,including the algorithms and further applications.The proposed two problems are different,however,they are also related,both of which have high theoretical values in the field of 3D modeling and biomedical engineering,and they have a wide range of applications in real life,such as animation design,pre-surgery evaluation and et al.Therefore,the study of these two problems also has practical value.In terms of local component analysis,this paper proposes a framework method for component segmentation of 3D models based on symmetric non-negative matrix factorization on dual graph.We design a similarity matrix to describe the affinity between two facets of the concerned model.Then,we utilize the achieved hidden variables for segmentation and saliency detection.Extensive experiments have verified the effectiveness and robustness of the proposed algorithm.In terms of global shape modeling,we propose a suite of methods for the overall atlas analysis of the deformable model.With data collection,data preprocessing,registration and segmentation of medical images,we study the commonness and difference among the data by principle component analysis.Then,we construct the deformable shape model of the Chinese human spine,and exploring more applications such as registration and Osteoporosis detection etc.Finally,we realize the above process through IDL language.To demonstrate the effectiveness of the method we propose in this paper,we adjust the parameters to observe the deformability,test the registration ability of the model by the leave-one-out method and pass the quantitative test of 5 osteoporosis patients which has been diagnosed.