Research On The Method Of Lung Surface Texture Synthesis And Mapping In The Virtual Surgery

The virtual surgery system is widely used in medical filed.It is the combination of virtual reality technology and computer graphics,which can provide medical staff with operation plan and operation training,etc.Lung is the important organ of human body.Improving the reality of lung is an important part of research on the virtual surgery system.The traditional two-dimensional texture synthesis algorithm mainly focuses on the structural texture to research,while the texture of human lung surface is generally both structural and random,and due to different lesions or differences in personal living environment and living habits,the texture of lung surface becomes more complicated.Mapping 2D textures to closed and irregular 3D model surfaces is prone to texture distortion,texture seams,etc.,resulting in irreality of the model.In view of these problems and the actual needs of virtual surgery,this paper establishes the texture mapping general framework of virtual surgery,and combining the characteristics of lungs,having conducted in-depth research on the synthesis and mapping of lung surface texture.The main work and innovation of the thesis includes:(1)Global lung surface texture synthesis based on Generative Adversarial Networks(GAN).The bilateral filtering algorithm is used to perform highlights removal for texture samples.This paper analyzes the characteristics of lung surface texture,and proposes a method for synthesizing lung surface texture based on GAN.The network generation model adopts a full convolution structure design,and introduces a attention mechanism in the residual module,which is based on the channel dimension and spatial dimension of the feature map.The training of the network is to double the spatial extent of texture blocks extracted from a specific lung surface texture exemplar.The trained generation model can synthesize the lung surface texture image with arbitrary size and local features and global features.(2)Lung surface texture mapping based on segmentation optimization.This paper improves the parameterization implementation of the triangular mesh automatic segmentation algorithm,and proposes an energy function combining the triangle separation metric in the grid and the triangular distortion metric based on AMIPS(Advanced Most Isometric Parametrizations).Based on the idea of homotopy continuation,the homotopy mapping is constructed by Gaussian smoothing of the non-convex energy function,to approximate the objective function,and the NewtonRaphson method is used to obtain the optimal approximate solution.It realizes the parameterization of the volume mesh with less segmentation and low deformation,and the parameter domain grid unit is as unfolded as possible.(3)Texture fusion at seams.It shows the bilinear interpolation reconstruction representation of the parameter domain texture,clarifying the continuity measure of the texture at the joint in order to obtain the mathematical condition of the seamless texture space.Combine the mathematical condition with the energy function at the seam to obtain the set of texels satisfying the condition,and the texels achieve the fusion and smoothing of the seam texture.Finally,this paper designs and implements the texture mapping system of the lungs,achieving the integration of main module algorithm and visualization,giving the experimental results and analysis.The algorithm described in this paper is not only meaningful for the surface texture of the lung,but also has important guiding significance for the surface texture of other similar organs in virtual surgery.