摘要
面部肤色的提取容易受到光照干扰,因此采用改进参考白和稀疏网络(SNoW)相结合的彩色人脸检测方法。首先,在对光照变化较大的人脸图像进行肤色提取之前,在HSV颜色空间进行参考白和直方图均衡的融合以减小光照变化引起的色彩偏差,再将参考白光照补偿结果进行肤色模型的人脸信息提取;最后借助稀疏网络模型法计算分割提取出的人脸区块特征并进行人脸的正确分类。实验结果表明,该方法实现了光照变化剧烈的复杂场景下较优的肤色提取及人脸的准确检测和定位,满足了系统实时性需求。
Extraction of facial skin color is vulnerable to illumination variation.In order to solve such a problem,this paper proposes an improved colored face detection method that combines improved reference white and sparse network of winnows （SNoW）.Firstly,before the face skin color extraction from the image with large illumination variation,the reference white and histogram equalization are fused in HSV color space so as to reduce the color deviation caused by illumination variation.Then,the extraction of the candidate human face regions can be achieved from the reference white light compensation results using skin color model.Finally,the sparse network model method is used to calculate the features in face segmentation blocks and classifies human faces correctly.Experiment results show that the proposed method can achieve good skin color extraction effect and accurate face detection and localization in the complex scene with illumination variations,and also meets the requirement of real-time performance.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2014年第4期820-826,共7页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61273277
61203261)
山东省自然科学基金(ZR2011FM032
ZR2012FQ003)
教育部留学回国人员科研启动基金(20101174)
国家重点实验室开放基金(SKLRS-2010-MS13)
高校博士点基金(20130131110038)资助项目
关键词
参考白
稀疏网络
ADABOOST算法
人脸检测
reference white
sparse network of Winnows （SNoW）
Adaboost algorithm
face detection
