多特征连续自适应均值漂移人脸跟踪算法  

Multi-feature face tracking algorithm with continuously adaptive mean shift

作　　者：丁业兵[1] 李敬仕[1] 方国涛[1] 谭学琴[1] 

机构地区：[1]安徽邮电职业技术学院通信工程系,合肥230031

出　　处：《计算机应用》2014年第A02期276-279,共4页journal of Computer Applications

基　　金：安徽省高校优秀青年人才基金重点项目(2013SQRL121ZD);安徽邮电职业技术学院院级项目(YJ201403ZR)

摘　　要：传统连续自适应均值漂移人脸跟踪算法,仅使用了人脸的色调特征,容易受到光照及相同背景色影响,为此,提出将人脸颜色和表示线端、角点、边界相关的纹理信息相结合共同构成人脸特征的CAMSHIFT算法。首先,采用带有权重的RGB颜色空间创建归一化的人脸直方图模型;然后,在视频图像中进行目标人脸直方图投影,同时进行纹理检测,保留纹理信息的人脸颜色特征,形成概率密度分布图;最后,由Mean Shift算法从当前位置迭代寻找直至定位目标人脸。实验结果验证了该算法可以抵御同色背景和部分遮挡的影响。Traditional Continuously Adaptive Mean Shift（ CAMSHIFT） face tracking algorithm only use hue characteristics of the face, while easily affected by illumination and the same color background. So, face color was combined with texture information that related to line ends, corners, edges to construct facial feature, in the CAMSHIFT face tracking algorithm. Firstly, the normalized face histogram model was created with weighted RGB color space; then the tracking face histogram was projected, and the texture in the subsequent video frame image was detected, so that the probability density distribution image with face color features retaining the texture information was formed; finally, the Mean Shift algorithm iterated looking up and the target face from the current position. The experimental results show that the algorithm can withstand the effects of the same color as the background and partial occlusion.

关 键 词：人脸跟踪 连续自适应均值漂移 概率密度 纹理 直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]