Research On Generation Method Of Example-based Facial Expression Animation

The diversity and uniqueness of the human face have been the hot topics in thefield of computer vision, graphics and pattern recognition. Since the uniquemanifestation mode, the cartooning art leads the new trend in the development of art.By the line depicted and color rendering, face cartoon can vividly represent the faceimage, and can be widely used in many applications, such as the online games,interactive BBS, social softwares and animation, etc.The existing face cartoon synthetic methods based on the sample learningalgorithm, usually divide images into small patches, and then use patch matching andsynthesis to generate the cartoon image. Due to the block-divided interferences, thesemethods cannot detailed describe the local facial features, which results in theuncomfortable results, especially the lack of the clear description of the sketch. Recently,the method has been proposed, which can achieve better sketch depiction by preciselylocating local features of the face.This thesis focuses on automatically generating the cartoon image with the artistpainting style form a given face image. Moreover, we also consider the changing of theexpressions on the cartoon face. The work of this thesis is introduced as follows:1. We proposed two face cartoon synthetic methods such as the parameter modeland the feature point method. Based on the active appearance model, the first algorithmutilizes the parameter estimation method to generate the cartoon image through learningthe process of the face matching. From the perspective of high level semantics, thesecond cartoon synthetic algorithm, which is based on the feature points, classifies thefacial features into three different parts and processes the hair, the facial contours andthe facial organs respectively.2. To achieve a colored performance, we proposed three methods to colorize thecartoon image, such as the color space conversion method and the image segmentationmethod. The cartoon image obtained by the color space conversion is similar to theinput image. However, the coloring image is closer to the artist’s painting style by theimage segmentation coloring method. 3. Some facial expressions of the generated cartoon image are done to enhancethe cartoon performance. Using the image warping algorithm, the expressionlesscartoon face can be changed into the smiling and sad face which is vivid and humorous.The experimental results demonstrate that the proposed face cartoon syntheticmethod can get comfortable performance, and the facial expression can make thecartoon much vivider with small time-consuming.