Research On3D Mesh Topological Perception Model And Its Applications

With the fast development and widely spread of3D scanner and interactive modelingtools, digital geometry with the expression of3D models, as the fourth generation digitalmedia is widely used and becoming and is more suitable for human visual system(HVS)and thinking pattern for their abundant visually perception details and contents than2Ddigital image. The increasing size of3D digital geometry data make it a indeed difcultthing to perform real-time description and rendering of a virtual scene. The visual per-ception based3D object processing theory and technology can significantly improve therender performance and visual experience of a3D scene under the condition of the samevisual quality and has become a eye-catching topic. They provide an efcient tool for3Ddigital geometry processing and application.The kernel of visual perceptual processing theory aims to provide a visual perceptionbased computational model, it mainly focuses on the following three aspects: visual atten-tion, visual masking and topology perception. They are in the form of3D mesh saliencymap, roughness map and sekelton respectively. However, existing research works fail toconsider the overall visual perception characteristics and lack of general visual perceptionmodel with corresponding objective visual quality and degradation evaluation methods.Visual meaningful3D mesh segmentation is an important part of visual perception theory,3D model segmentation based on Markov Random Field(MRF)provides an efcient mod-eling tool for integrating the geometric features with visual prior knowledge. However,visually meaningful based3D mesh segmentation using MRF is in its beginning stage andshort of a MRF based segmentation framework having the capability to integrate diferentfeatures. Aiming at the above issues in reserarch hot topics, the main research works andcontributions of this dissertation are as the following:(1) A general visual perception based computational model–SMTPM(Skeleton Mod-ulated Topological Perception Map) is proposed. Considering the attention mechanism,masking mechanism and global topological perception of HVS, an improved bottom-upattention oriented saliency map model is presented. A global topology perception orient-ed and kinematics importance guided skeleton map algorithm is also proposed. Based onmodulation between saliency and roughness maps using the skeleton map, We propose a general visual perception model which can describe the attention and masking mecha-nisms of HVS under diferent degradation status.(2) A multi-scale visual degradation evaluation algorithm based on SMTPM is p-resented and used to compare mesh simplification methods. Considering the situationof SMTPMs′degeneration to saliency map, a global perceptual degradation evaluationmethod based on Shannons′information entropy is proposed and used to evaluate the vi-sual degradation during mesh simplification and compare mesh simplification methods.The subjective corpus and subjective experiment protocol are also designed. The statisti-cal analysis results for conducted subjective experiments demonstrate that the subjectiveexperiment protocol is valid and the evaluation method is superior to other geometricbased metrics. The capability of multi-scale visual degradation evaluation is also verified.(3) A rapid viewpoint selection method and a definition of perceptual entropy areproposed. Based on SMTPM, this paper presents the definitions for the amount of view-point perceptual information. The best and worst viewpoint selection methods based onsubdivision stencil are also proposed. Additionally, a Spherical Viewpoint PerceptualInformation Map(SVPIM) is defined based on the amount of viewpointsp′erceptual in-formation. SVPIM provids a uniform expression for visual perceptual information ofarbitrary3D objects. The perceptual entropy is first proposed using SVPIM and provide atheortical guide and general visual perception quality evaluation for3D object processing.(4) Based on the skeleton map and a unified segmentation framework using MRFpresented in this paper, a new visual meaningful mesh segementation is achieved. MRFbased clustering is the kernel of this framework and any3D mesh item with geometric orperceptual features can be easily integrated into the framework. A new visual meaningfulmesh segmentation algorithm is implementated by introducing the visual prior knowledgeto the framework and utilizing the double level Gibbs random field. Experimental result-s show that segmentation algorithms based on the proposed framework achieves bettersegmentation results than other clustering based segmentation algorithms.