基于轮廓信息的人脸检测

针对彩色图像提出了一种基于肤色模型、脸部轮廓信息以及眼睛特征的人脸检测算法。采用基于YCbCr色彩空间的肤色分割模型，初步筛选人脸的候选区域；在此基础上进行边缘检测，获得人脸轮廓信息，并利用遗传算法拟合脸部的椭圆；在椭圆的水平方向根据眼睛的几何特征来检测“眼睛对”，再根据“三停五眼”来定位人脸，并利用左右对称性验证人脸。实验表明，该算法对于彩色图像的正面人脸检测具有良好的效果。

Face Detection Based on Contour Information

One novel algorithm for face detection in color images is proposed,which is based on skin model,face contour information and geometric features of human eye.First of all,a special skin model related with YCbCr space is employed to locate candidate faces in color images containing faces.Then,contour information of human faces are computed by means of Sobel operator,and the match between ellipse template and face contour was optimized by GA in the edge energy field of face image.Thus if the ellipse corresponds to a real face would be determined by if an eye-pair exists in this ellipse.Finally,all regions of real faces could then be obtained according to the concept that a standard face could be partitioned equally into three parts vertically and five parts horizontally.Experiments show that this method achieves satisfactory performance in face location and detection.