A Rapid Costume Textures Generating Method Of Virtual Human Characters

In a large-scale crowd simulation scenario,using a single repeated costume texture for virtual human characters will reduce the credibility of the simulation scene,while generating virtual characters with complex costume textures is costly.Therefore,this thesis studies a method for quickly generating virtual characters with costume textures.Firstly,using algorithms to parse different parts of the costumes from the existing character pictures,then performs the texture mapping and texture map's generation to realize the direct generation of the character costumes' texture from a single image.The main work of this dissertation is as follows:1)The current research status and achievements of generating three-dimensional virtual character costume models from pictures are reviewed.The existing systems and algorithms used in generating similar human character with its costume textures from three-dimensional pictures and the algorithms of parsing character's costume parts from the pictures are analyzed and summarized..2)Combining deep learning and GrabCut algorithm to realize the analysis and extraction of different costume parts from character pictures.For the pictures of modern characters,the deep learning method macro-micro Adversarial Network(MMAN)is effective with the advantage of convenient operation.Nevertheless,for the pictures that the deep learning method performs poor recognition effects,the interactive GrabCut algorithm is introduced to improve these pictures' recognizing accuracy.3)A method for quickly generating virtual human character with costume texture is proposed.Firstly,the algorithm mentioned above is used to parse and extract the costumes.Then,each extracted costume parts is mapped with its corresponding locations on the three-dimensional character model texture map by the algorithm.As for the occlusion part of the costumes in the pictures,the clothing color style information is calculated as the supplements.Finally,the 3D human character model is obtained through the 3D character costume texture map which retargeted from a single picture.4)Design related experiments and complete the generation of virtual human character model with costume textures from a single image.The experiment results verify the effectiveness and convenience of the proposed algorithm.