Face Portrait Stylized Rendering Technology Research

Non-Photorealistic Rendering (NPR) injects aesthetic and stylized elements into computer graphics. NPR primarily aims to conveying not only compressed information but also a certain artistic spirit, by simulating various artistic styles such as pencil sketching, watercolor painting, or Chinese ink painting. And the results can be applied to the advertisement propaganda, interior decoration, cartoon making and others areas.Face portrait stylize rendering is a branch of the non-photorealistic rendering research. This thesis focuses on the research of the background, significance and relevant key skills of portrait stylize rendering research, and finally give out the sketch of the experiment result.Firstly, this thesis presents with a generation algorithm of face portrait for Yunnan heavy color painting by researching Yunnan color painting style represented by Ding Shaoguang, and transforms the face images into heavy color painting style. As a very important part of Yunnan heavy color stylized rendering research, results of this algorithm will affect the whole painting effect. In the portrait generation algorithm of Yunnan heavy color painting, face drawing rules of Yunnan heavy color painting is researched, and moreover features of deformation and extraction of line graph of Yunnan heavy color painting style is realized by Moving Least Square Method and Flow-based Difference-of-Gaussians (FDOG). The algorithm can effectively simulate the styled rendering of Yunnan heavy color painting and achieved some painting effect.Secondly, this thesis deals with the study of cartoon rendering technique of NPR, too. First of all, the input image is abstracted via bilateral filter, which boosted up the contrast. Then abstracted images can optionally be stylized by using soft color quantization to create cartoon-like effects, and edge extraction based on FDOG is used to improve the visibility of the cartoons. Finally, the quantized pictures are fused with edge images to get the cartoon style effect with color, higher visibility and fewer details.Finally, aiming at the problem that quantization interval between the obvious transition discontinuities in the image quantization process, this thesis uses soft quantization to improve the effect. The luminance gradient function is introduced to allow hard boundaries only where the luminance gradient is high; and in low gradient regions, boundaries are spread out over a larger area. And finally, a quantized sketch with bright color and continuous quantized interval can be obtained.