基于肤色正态分布模型的自适应人脸动态检测

Adaptive face detection based on normal distribution model of skin color

为了实现不同类型人脸的自适应检测,并提高人脸检测精度和效率,利用色度空间中肤色分布的良好聚类性,提出了基于肤色正态分布模型的自适应人脸动态检测算法.文中首先在人脸图像亮度补偿基础上,建立了YCb Cr颜色空间中的肤色正态分布模型,并利用该模型计算,得到像素点属于肤色区域的概率,即两者相似度;然后对肤色相似度进行中值滤波和灰度化;最后实现了人脸的快速自适应阈值分割和定位.实验结果表明:在保证不同脸型的高检测率和低误检率同时,该检测算法的定位精度和实时性也得到了进一步提高.

To realize adaptive detection of different types of faces,and improve accuracy and efficiency of face detection,using good clustering of skin color distribution in color space,an adaptive face detection algorithm is presented based on normal distribution model of skin color.First,on the basis of luminance compensation of a face picture,a normal distribution model of skin color is built in YCbCr color space,and the probability that face pixels belong to the zone of skin color （namely the similarity between them）is calculated through the model. Then the skin similarity is subjected to median filter and gray processing.Finally,a face is quickly segmented through an adaptive threshold method and located.The experimental results show that the proposed algorithm im-proves the detection rate and reduces the error detection rate of different types of faces;furthermore,its location accuracy and real-time performance are further improved.