彩色视频图像卡通风格化研究与实现

提出一种彩色视频图像卡通风格化的新方法,将均值漂移滤波与高斯图像金字塔结合起来使用,在对彩色视频图像进行颜色聚类和噪声消除的同时,提高了处理的速度.运用DoG算子对滤波后的图像进行边缘检测;再对移滤波后的图像进行色彩量化;最后将量化后的图像与边缘图像进行融合.融合时根据边缘处梯度的大小设置边缘颜色的深度,使得最终得到的卡通化图像不仅在高对比度区域得到加强而且减少了噪声,边缘更加自然.实验结果表明:本算法与传统的双边滤波和DoG算子相比,获得了更好的卡通风格化效果和更快的运行速度,并且可以直接将输入的视频文件转换为卡通化的视频文件输出.

Research and Implementation of Cartoon-Like Stylization for Color Video Image

A new approach for cartoon-like stylization of color video image is proposed. The Mean-shift filter and gaussian pyramid are combined in this approach to cluster the color and eliminate the noise of the color video image and improve the speed of processing at the same time. Then, the DoG is used to detect edges of the image which has been filtered by Mean-shift filter. After that, the soft quantization is applied to the image which has been filtered by Mean-shift filter. Finally, the edge curves and the quantitative image are fused to generate a personalized cartoon image. In this step, the color of edges in the cartoon image is based on the edge gradient. This makes the cartoon image not only be enhanced in the regions of high contrast, but also have less noise and seems more natural. In comparison with the approach for cartoon-like stylization of image based on the bilateral filter and DoG, the experimental results show that our approach can produce a better cartoon effect in shorter time, and can directly convert an input video to a cartoonish video as an output file.