摘要
人脸识别中,发型遮挡是一种十分常见的遮挡类型,并且对人脸的正确识别具有极大的干扰。提出一种将头发的颜色模型和发型特征相结合的遮挡检测方法。首先,采用机器学习的方法,对头发的颜色进行学习建模。然后,利用发际线的特征,将人脸划分为若干扇形并分块,采用逐步精细的方法对人脸的发型遮挡区域进行检测。实验结果表明,该方法对人脸区域发型遮挡检测的准确率和召回率都达到88%以上,相比PCA检测方法提高了约20%,验证了该方法的有效性。
In face recognition,the hair is a very common object in face occlusion,and it has a great disturbance to the correct recognition of human face.In this paper,a method of hair-occlusion detection is proposed,which is based on the combination of color model and hair style.First of all,using machine learning method to learn the color of hair and build the hair-color model.Then,the face is divided into several segments by using the feature of the hair line,and then the hair-occlusion area is detected by the method of fine precision.Experimental results show that the accuracy and recall rate of the proposed method is more than 88%,compared to the PCA detection method increased by about 20%.
出处
《微型机与应用》
2016年第2期32-34,共3页
Microcomputer & Its Applications
基金
北京市属高等学校创新团队建设与教师职业发展计划基金项目(IDHT20130519)
北京市学科与研究生教育基金(PXM2015_014224.000018)
关键词
人脸识别
发型遮挡
遮挡检测
face recognition
hair occlusion
occlusion detection