针对人脸检测过程中难以区分人脸与非人脸等问题,提出了一种基于级联Adaboost 和神经网络主元分析(PCA)算法的人脸检测新方法以提高人脸检测的正确率.采用两级检测器对人脸进行区分检测:首先将计算速度较快的Adaboost算法作为第一级检测器对人脸图像快速扫描,对所有判断为人脸的窗口进行合并.然后将合并的窗口提取特征并送入作为第二级检测器的PCA进行验证,排除那些不可能是人脸模式的窗口.最后经过PCA检测结果判别输出验证后的人脸窗口参数(包括窗口的大小和位置信息).不同算法检测结果显示,基于本方法的人脸检测正确率达到了92.6%,检测率为94.1%;基于Adaboost检测正确率为62.5%,检测率为88%;基于SVM检测正确率为54%,检测率为89%;基于FSS 检测正确率为66%,检测率为92%.实验结果表明,本方法能够很好地区分人脸模式和非人脸模式.因此,级联Adaboost和PCA算法组成的两级检测器可以明显提高人脸检测系统的性能.%For it difficult to distinguish faces and no-faces in the face detection, a new method was proposed based on the algorithm of Adaboost and PCA that can improve the accuracy of face detection.The method uses two-stage detector to detect faces,first it takes the algorithm of Adaboost as primary detector scan pictures of face quick-ly and merge all windows of face that be judged as face,then verify the correctness according to sending window′s features to secondary detector,that can excludes the windows of no-face,last verify the exporting parameters of face which includes the size and position of the window.Different algorithm test results show the accuracy of face detec-tion based on this method was 92.6%,and the detection rate was 94.1%;the accuracy of based on the Adaboost was 62.5%,which was 88%;the accuracy of based on SVM was 54%,which was 89%;the accuracy of based on FSS was 66%,which was 92%.The experiment result shows that this method can distinguish human face pattern and non-face pattern.So in that sense,a two-stage detector of the cascade′s Adaboost and PCA algorithms can sig-nificantly improve the performance of the face detection system.
展开▼
