摘要
为了在提高复杂背景下的人脸检测率的同时减少检测时间,将肤色分割和Haar方差特征相结合,在YCbCr颜色空间通过椭圆肤色模型和logistic回归分析确定每一点的肤色概率,生成肤色概率图,从而将每一点的像素值映射到[0,1],在Ostu方法的基础上采用并行的遗传算法确定肤色分割的阈值,快速分割出人脸区域;最后用少量的Haar方差特征取代原来的Haar特征,并采用SVM训练分类方法对分割出的人脸区域进行验证。实验表明,该方法不仅提高了人脸检测的正确率,而且具有较快的人脸检测速度。
In order to improve the rate of face detection under complex background and reduce detection time,we propose a face detection algorithm based on skin color segmentation and improved Haar features.First we derive the probability of every bit of skin color by elliptical skin color model and logistic regression analysis in YCbCr color space and generate the skin probability diagram,which will map every bit of pixel values to [0,1].Then we define the threshold of skin color segmentation to segment the face region rapidly,using parallel genetic algorithm based the Ostu method.Finally combining the SVM training and classification method,we adopt a few Haar variance characteristics instead of the original Haar features to verify the segmented face region.Experimental results show that the proposed method can improve the accuracy of face detection with a fast speed.
出处
《计算机工程与科学》
CSCD
北大核心
2015年第1期146-151,共6页
Computer Engineering & Science
关键词
肤色分割
椭圆肤色模型
回归分析
遗传算法
方差特征
skin color segmentation
elliptical skin color model
regression analysis
genetic algorithm
Haar variance characteristics