卡通化图像和视频的毛发纹理生成与替换

Cartoon Fur Texture Synthesis and Replacement in Images and Videos

低成本卡通制作中的图像和视频通常缺乏对动物角色毛发效果的表现,为了能对已有图像及视频中的动物角色进行处理,为其增添具备真实感的毛发效果,提出一种毛发风格化算法——卡通化毛发纹理算法.针对卡通中的动物角色合成毛发纹理并进行替换,分为图像应用及视频应用2个部分.在图像替换时,对要进行风格化处理的目标区域进行图像结构分析,以获取覆盖目标区域的三角网格,再生成毛发纹元并映射于网格之上,通过绘制纹元来生成具备真实感的毛发效果;在进行视频替换时,提取视频关键帧并基于图像应用算法生成相应的卡通化毛发纹理进行图像替换,之后根据关键帧的替换结果指导整个视频的替换.为了获取随时间变化的图像关键帧目标区域,采用SIFT算法计算特征点在时间轴上的匹配;为了快速合成卡通化毛发纹理,采用基于GPU的光线行进算法加速毛发纹元的体绘制过程.实验结果表明,文中算法可成功地对已有图像及视频的动物角色添加具备真实感的毛发效果.

Low-cost cartoon images and videos often lack the representation of furry effect for animal characters. In order to create image or video with complex furry stylized texture from the original low- cost one, we present a novel furry stylization method, called as cartoon fur texture technique, to allow furry stylized textures to be generated correspondingly for the given image or video. By the method, on the one hand for the image application, the triangle mesh for the ROI（Region of Interest） area in the image can be obtained firstly by analyzing the input image structure, then the fur texel can be mapped onto the mesh and rendered to achieve the realistic furry result. On the other hand, the furry stylized texture synthesis on image can be applied to each image as a basic algorithm for video application, with the key frames of the video calculated firstly and then the furry stylized texture in each key frame synthesized, to generate the cartoon fur texture. In our experiments, in order to calculate the feature matching of ROI along with the timeline conveniently, the SIFT（scale invariant feature transform） algorithm is applied. Besides, we use the Ray Marching technique on GPU to implement the volume rendering of texel in a much improved speed. Experimental results show that the presented method can be used to successfully synthesize the furry stylized textures for source images and videos, with realistic furry effect well achieved.