基于双高斯平均似然度和帧间差分的人脸视频图像肤色提取

摘要

For facial video imags with colored lighting and complex background, a skin color extraction algorithm is presented based on the CbCr-CgCr double Gaussian average degree of likelihood and the inter-frame difference of component Ⅰ in YIQ color space. The algorithm calculates the CbCr-CgCr double Gaussian average likelihood of facial video input image after preprocessing, segments the average likelihood image roughly by using the value of average likelihood and component Ⅰ, and extracts the skin color region by combinating the skin coarse segmented image and component J inter-frame difference image. Experimental result demonstrates that the algorithm can obtain good results in skin color extraction.%针对有色光照和复杂背景的人脸视频图像,提出一种基于CbCr-CgCr双高斯平均似然度和YIQ颜色空间的Ⅰ分量帧间差分的肤色提取算法.该算法对色偏调整预处理后的人脸视频输入图,计算CbCr-CgCr双高斯平均似然度,并利用平均似然度值和Ⅰ分量对平均似然图进行粗分割,然后结合平均似然度分割图及Ⅰ分量帧间差分图,提取图像肤色区域.实验证明该算法有较好的肤色提取效果.