基于双色彩空间和面部矩形特征的人脸检测  

Face Detection Algorithm Based on Double Color Space and Facial Rectangular Characteristics

作　　者：朱志亮[1] 熊凤[1] 陶向阳[1,2] 刘晓山[1,2] 

机构地区：[1]江西师范大学物理与通信电子学院,江西南昌330022 [2]江西省光电子与通信重点实验室,江西南昌330022

出　　处：《江西师范大学学报（自然科学版）》2013年第4期367-370,386,共5页Journal of Jiangxi Normal University(Natural Science Edition)

基　　金：江西省教育厅科技项目(GJJ13210);江西省光电子与通信电子实验室基金(2011001);江西师范大学博士启动基金(4029)资助项目

摘　　要：在YIQ和YCbCr色彩空间中,结合人类面部矩形的特征,提出了一种用于人脸检测和定位的算法.该算法通过比较输入图像R、G、B 3通道总的平均值和B通道的平均值来选择色彩空间,并采用基于积分图的2维OSTU法进行肤色分割;然后根据肤色区域面积对肤色分割图像进行中值滤波;最后根据中值滤波后的肤色分割图中的肤色区域面积构造最佳人脸矩形,并用该矩形去定位人脸区域.实验结果表明:该算法较传统的采用单一色彩空间的人脸检测算法具有更高的准确率.An algorithm for face detection and location based on human facial rectangular characteristics in the YIQ and the YCbCr color space has been proposed.The algorithm selects the color space by comparing the average of the input image channels of R,G and B with the average of the channel of B,and segments the skin area with the twodimensional OSTU and integral figure.Then,the skin segmentation figure by median filter based on the acreage of the skin area has been processed.In the end,an optimal face region rectangle with the acreage of the skin area which has been processed by median filter has been constructed,and the rectangle to locate the face region has been used.The experiments show that the algorithm has higher accuracy than the traditional face detection algorithms using single color space.

关 键 词：色彩空间 肤色分割 积分图 OSTU算法 人脸检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]