基于关键标记点高频分析的模糊人脸图像鉴别

Method of Bluring Face Identification Based on the High Frequency Analysis of Key Landmarks

由图像模糊造成的人脸图像质量降低是直接影响人脸识别精度的重要因素之一.对模糊人脸图像进行鉴别能够为图像的采集及后期的处理识别提供指导意见,从而降低图像模糊对人脸识别的影响.本文提出了一种基于关键标记点高频分析的模糊人脸图像鉴别方法.基于图像模糊最直接地反映在图像细节信息上的特点,本文方法从人脸上最能体现细节信息的标记点入手,首先提取人脸关键标记点(眼睛、眉毛、鼻子、嘴的标记点),以每个关键标记点为中心取图像小区域作为特征分析对象;对每个特征区域进行高频信息分析,提取代表特征区域是模糊还是清晰的特征,并引入对应的分类器判断各特征区域是否模糊;最后,依据每个特征区域的判断结果,以投票的方式决定整幅人脸图像是否模糊.在公开数据集上进行了大量实验,验证了本文方法的有效性.

The deterioration of image quality caused by blurring is one of the most important factors effecting face recognition accuracy.The identification of blurring face can provide useful guidance for face image′s acquisition,processing and recognition,and thereby reduce the adverse impact caused by blurring on face recognition.In this paper,we propose a method of blurring face identification based on the high frequency analysis of key face landmarks.Since image details prominently reflect characteristics of blurring,our method is mainly designed focused on face landmarks which well reflect face details.Firstly,the method extracts location of key face landmarks （landmarks of eyes,brows,nose and mouth）,and a small square surrounding each landmark is seen as feature region.Secondly,the high frequency analysis is conducted to each feature region to extract representative classification features,and the bluring classifiers correspond to each feature region are designed to distinguish the feature region is blurring or clear.Thirdly,voting is employed to conclusively identify blurring face.A large number of experiments have been conducted on publicly available data set to verify the effectiveness of our method.