一种改进的人脸检测方法  

An Improved Face Detection Method

作　　者：杨定礼[1] 张宇林[1] 赵环宇[1] 季仁东[1,2] 刘保连[1] 白秋产[1] 

机构地区：[1]淮阴工学院电子与电气工程学院,江苏淮安223003 [2]南京航空航天大学理学院,江苏南京210016

出　　处：《实验室研究与探索》2015年第2期111-116,共6页Research and Exploration In Laboratory

基　　金：国家自然科学基金(61203056);淮安市工业项目(HAG2013064);淮阴工学院基金项目(HGB1202)

摘　　要：当光照和肤色变化较大时,肤色的色度值易受影响,因此提出了改进的高斯肤色模型方法,此方法只提取人眼下方的部分像素作为样本,获得了自适应的肤色的色度值。同时,由于图像矩阵奇异值分解以后,其维数较高,因此引入Frobenius范数来降维。降维以后,为了能够实现非线性可分,提高训练速度与人脸检测率,又提出了改进的决策树SVM分类方法。实验结果表明,用改进的高斯肤色模型与改进的决策树SVM分类的方法相结合不仅提高了人脸检测准确率,而且还降低了误检率与漏检率。When illumination or skin color change significantly, chromaticity value of the skin color is likely to be affected. Thus, an improved Gaussian skin color model is proposed. A part of pixel below two eyes is extracted in this method. And adaptive chromaticity value of the skin color is obtained. Simultaneously, after matrix of image is decomposed with SVD, its dimension is high. Thus, the Frobenius norm is used to reduce dimension. In order to realize nonlinear separability and increase speed of training and the rate of the face detection, an improved method based on decision tree SVM classification is proposed. Experimental results show that it can increase the rate of face detection, and decrease the false detection rate and false negative rate.

关 键 词：肤色 奇异值 FROBENIUS范数 决策树 SVM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]