摘要
提出了一种新的基于肤色的多人脸检测方法。该方法先通过肤色分割得到人脸候选区,然后结合图像的小波表示和主元分析方法通过训练得到可用于区分人脸和非人脸的特征向量,并用改进的贝叶斯分类器对输入图像进行多人脸检测,改进的判决准则中参数,可用于控制检测的准确率和虚警概率,通过设定不同值可使算法适用于不同要求的应用,另外为保证获得较高准确率的同时降低虚警概率,还提出在经分类器判决后的人脸区域中依据对应的马赛克模板进一步排除虚假人脸。
A new multiple face detection method based on skin is proposed, which got the face candidates by the help of skin color and makes use of wavelet express of the images and the principal component analysis （PCA） to get the eigenvectors distinguishing faces and non-faces, and modified Bayes classifier to detect multiple faces ofinput images, ω, the parameter ofthe modified rules, could control detection accuracy and error probability to apply to different application by setting the values of ω. In addition, after classing, a mosaic template is used to exclude fake faces to ensure high accuracy and low error probability.
出处
《计算机工程与设计》
CSCD
北大核心
2008年第9期2304-2306,2349,共4页
Computer Engineering and Design
关键词
人脸检测
肤色分割
主元分析
贝叶斯准则
马赛克模板
face detection
skin color segmentation
PCA method
Bayesian rule
mosaic template
