Three-Dimensional Human Modeling For Virtual Dressing

Online shopping industry is developing continuously. As the leader Of online shopping industry, Online apparel shopping industry is growing prosperity, but it also face with a series of problems. These problems stemmed from the contradictions between "the virtual activity " (online shopping) and "physical entities" (clothing). The Three-Dimensional Virtual Fitting System will fundamentally solve those problem. Three-Dimensional human modeling is an important part of the Three-Dimensional Virtual Fitting System, which will not only make a very positive impact on E-commerce development, but also has very wide range of application. The thesis contains the following three parts:The research goal is to match requirements from online apparel shopping industry, we would like to design and build a three-dimensional virtual human modeling system, we focus on the improved algorithms of the human contour extraction, feature points extraction, three-dimensional human modeling, we try our best to minimize time and resource consuming in each step in the premise of modeling accuracy. In this way, we achieve a faster, lower cost human modeling system which meets application requirements.The main contents of this thesis contains the following three parts:Firstly, contour extraction algorithm based on the Two-Dimensional image of human body. Different from other active contour models, we use Chan-Vese model. Chan-Vese model is a region-based model, it does not depend on the image edge, neither subject to the position of initial contour, nor the shape of that, The results of the extracted contour are always accurate. Therefore, this thesis applies the Chan-Vese model to contour extraction based on Two Dimensional image of human body, the extraction results was also satisfactory.Secondly, feature points extraction algorithm based on the body contour. The purpose of the feature point extraction is to create the correspondence between two-dimension contour of real human body and three-dimensional contour of virtual human model. We set up mathematical model to finish the key features definition and extraction, which contains key points and auxiliary points, we not only successfully extract bump points, but also successfully extract concave points. Besides, our algorithm can automatically extract all feature points, and avoid human-computer interaction.Thirdly, three-dimensional human modeling. This view-dependent deformation technology deform the standard human model to generate the target human model, according to the correspondence between feature points in two-dimensional image of template body and those in three-dimensional template human model, the correspondence between feature points in two-dimensional image of target body and those in three-dimensional target human model. This technology only allows the vertex moves along the direction which perpendicular to the view direction, we also introduce contribution function, the measures solve the problem of local intersection, and also achieve local continuity during deformation.Finally, we make analysis on three-dimensional human model experimental, which prove that the superiority of our proposed algorithm for three-dimensional human modeling, and it can be applied to virtual fitting system.