摘要
介绍一种基于DirectShow的人脸快速检测系统的实现。首先利用DirectShow获得需要处理的视频帧,构造了皮肤颜色模型,对待测图像中肤色区域进行分割,检测出疑似人脸区域,并计算出该区域相对于皮肤的马氏距离图;根据马氏距离图寻找特征脸子空间,构造出BP神经网络;以马氏距离图在各特征脸子空间的投影系数为BP神经网络的输入,计算检测出的区域是否是人脸图像。实验证明,该系统具有良好的效果。
A face detecting system with DirectShow technique is presented. At first, the video frames are obtained by using DirectShow technique. And then, the skin-scolor model is constructed, which can detect the suspected skin areas, and calculate the map of Mahalanobis distance of these areas relative to skin. Based on the map of Mahalanobis distance, the eigenface subspace is found out and a BP neural network model is constructed, On the BP neural network, the input is coefficient vector obtained from map of Mahalanobis distance projecting on each subspace, and the output determines whether the detected area is face image or not. The experimental results show that this system provides good performance.
出处
《电视技术》
北大核心
2008年第8期88-90,共3页
Video Engineering
