The research on Photorealistic Face Texturing has both significant academic importance and wide application in Computer Graphics. With the aim of generating photorealistic face texture, we focus our research on several key techniques in this field, including valid constrained parameterization and multi-view texture blending, we apply and improve the Face Alignment technique-Active Appearance Models (AAMs) in our system. Based on this , we can fast and accurately locate the constrained dots in multi-view images. We realize a photorealistic face texture generation system with simple and fast operation. Combined with personal head modeling, we can produce a photorealistic personal head model, which can be used in computer games and animation. This thesis’ main contribution includes: (1) We proposed a combined feature of gray, edgeness and gradients from two directions into Active Appearance Models, greatly improving the precision of the face alignment. We also applied the improved AAMs in side face alignment, and proved that the result satisfied the high precision requirement in face texturing application. (2) With regard to multi-view texture mapping, we made improvement from three aspects: We proposed a discretization method, making the parameterization to be pixels, which can richly used the texture information in multi-view images. We proposed a search method based on the 3D topology and the barycentric coordinate during the process of locating a pixel in some triangle, and this method worked efficiently. We proposed a multi-view weight setting method combining the 3D triangle normal and the images information together, which makes the texture blending more smoothly and naturally. Our photorealistic face texturing method needs no calibration registration, and has wide application. We have already applied it in personal human body modeling.
修改评论