Research On 3D Personalized Face Reconstruction Based On RBFs

Realistic face modeling is one of the most difficult problems in computer graphics. With the development of virtual reality, character animation for films and advertising, computer games, realistic face modeling attracts many researchers' attention. 3D personalized facial reconstruction has very important theoretical and practical value for fast creation of personalized face model which is the basis of realization of realistic huaman animation.In this thesis, we firstly introduce the applications and present situation of 3D face modeling research at home and abroad, then analyze and summarize the main existing face modeling methods. On the basis of above, we present the method of creation of generic facial model with regular mesh by hierarchical design, the method of facial feature extraction with facial feature templates, the method of 3D personalized face reconstruction based on multi-region with RBFs, the method of approximate view-independent facial texture generation and mapping, which are described as bellow.(1) The method of creation of generic facial model with regular mesh by hierarchical design is presented. According to the structure of human face, we create a genenic facial model with regular mesh by hierarchical design from a raw facial model which is exported from Poser. This generic facial model has simple structure and good reference for construction, which laies good foundation of personalized facial reconstruction.(2) The method of facial feature extraction with facial feature templates is presented. Due to the feature in front and profile facial images isn't bijection, 3D feature coordinates can't be generated from facial images. We use feature templates to extract facial feature from images, which is convenient to calculate 3D facial feature coordinates from facial images.(3) The method of 3D personalized face reconstruction based on multi-region with RBFs is presented. Due to the complex structure of human face, the methods which use a constrained equation to construct 3D personalized face can't delicately represent facial organ surfaces, and the computation cost of these methods is greater. We present a method for reconstructing 3D personalized face based on multi-layer and multi-region with RBFs. Experiments show that this method is simple and efficient.(4) The method of approximate view-independent facial texture generation and mapping is presented. We analyze the disadvantage of traditional view-independent facial texture generation method, and present an approximate view-independent facial texture generation and mapping method. Experiments show that this method is simple and the personalized facial model textured is realistic.(5) A prototype system of 3D personalized face reconstruction based on front and profile facial images is designed and implemented by using the object-oriented methods. By running test of this system, we can conclude that the above methods are effective and the reconstructed personalized face is realistic.