面向在线实时应用的卡通风格化方法  

Cartoon-like Video Stylization for Online Real-time Applications

作　　者：洪朝群[1,2] 陈旭辉[1,2] 

机构地区：[1]厦门理工学院计算机与信息工程学院,福建厦门361024 [2]福建省高校物联网应用技术重点实验室,福建厦门361024

出　　处：《厦门理工学院学报》2015年第1期73-78,共6页Journal of Xiamen University of Technology

基　　金：国家自然科学基金项目(61202145);福建省自然科学基金项目(YKJ12005R)

摘　　要：针对视频的卡通风格化存在着处理速度慢和连续帧中边缘视觉效果差的问题,提出具有高实时性的视频卡通风格化处理方法.该方法主要包括扩展非线性扩散、边缘检测、彩色量化和融合4个步骤.边缘检测算法基于高斯-拉普拉斯算子进行改进,速度更快而且在连续帧中视觉效果更好;在扩展非线性扩散和色彩量化的过程中利用查表法获取滤波系数和量化值,避免了对每一个像素计算高斯方程和量化方程,进一步提高了速度;同时,在传输过程中对边缘图像和量化图像采取分离编码进行传输,降低了码率.实验证明,在视频聊天、在线游戏等实时性要求高的应用场合中,该方法对人脸和景物都取得了良好的卡通风格化效果.Cartoon-like stylization for video suffers the problems of low speed and unpleasant visual effect of edges in contiguous frames. In this paper, a novel approach of cartoon-like stylization for video is presented, which works well for occasions with a high real-time requirement. Generally speaking, this approach consists of four steps： extended nonlinear diffusion, edge detection, color quantization and fusion. In this approach, the edge detection algorithm is improved based on the Laplacian of Gaussian （LOG）. It is faster and has better visual effect of edges in contiguous frames. In addition to that, an implementation based on querying the tables of filter coefficients and quantization values is used in extended nonlinear diffusion and color quantization. It avoids computing Gaussian function and quantization function for each pixel and further improves the speed of the proposed method. Furthermore, the encoding process of the edge picture and quantized picture are separated, which reduces the bit-rate. The experimental results show that the proposed method performs well on the human faces and scenery in the circumstances of video chatting and online games.

关 键 词：卡通风格化 实时应用 边缘检测 扩展非线性扩散 色彩量化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]