Machine Learning-based Face Cartoon

The research of face image has been a popular research direction in the field of machine vision, pattern recognition and computer graphics, which is widely used in video conferencing, network game and so on. In recent years, automatically generating the cartoon image based on the face image becomes one of the researches of hotspots at home and abroad. It is difficult to generate appearance-sprite in one cartoon image automatically.The main work of this paper is to research the algorithm of automatically generating cartoon image from the face image; its goal is to generate image lifelike cartoon image. This algorithm is based on Markov Network model. First, we build a set of face-cartoon patch. The input image is divided into overlapped patches in the same size, for each photo patch can find several similar photo patches in the photo patch set, which its corresponding cartoon patches are the candidates of the input photo patch. Then we build Markov Network model and propose the idea of choosing compatibility, which can simplify the solving process of Markov Network. At last, it makes use of the method of combining direct average stitching technique and bilateral filtering and output the cartoon image.The main research is as follows:1. According to the nonlinear relation between face and cartoon, this paper researched the algorithm of generating cartoon face, which is based on patches of Markov Network model. It used limit matching technology in the searching process, which can improve the semantic relevance among patch matching and reduce the search complexity at the same time.2. Since it is intractable to get in the Markov Network model, we propose the thought of that the overlapping area between neighboring patches has the greatest compatibility in horizontal direction. It simplifies the solving process, and speeds up the Markov Network convergence. Since the size of each cartoon patch is small and contains limited amount of information relatively. So we apply the method of combining direct average stitching technique and bilateral filtering in the process of synthesis. It generates a colored cartoon image by the method of color space transformation with strong entertainment.It has shown that the proposed algorithm has the characteristics of higher similarity and fast output from the experimental results. It has strong practical value.