基于深度学习的低质量图像模糊人脸识别方法

A method on fuzzy face recognition in low-quality image based on deep learning

为避免灰度区域含噪度对人脸识别准确度的影响,提出基于深度学习的低质量图像模糊人脸识别方法.采用小波变换离散化低质量图像,获取包含不同方向特征的轮廓线尺度信息.根据人脸面部特征,检测人脸轮廓线关键点,并通过计算待识别样本与源样本的协方差矩阵、图像的隶属度函数,对噪点进行均匀采样,以降低含噪度.结合关键点阈值函数,多角度匹配关键点.采用深度学习算法构建识别分类器,并计算多角度匹配关键点结果中两个样本之间的距离实现人脸识别.测试结果表明,在不同模糊比例的人脸图像识别中,所提方法得到的识别准确度均较高,证明所提方法可以准确识别低质量图像模糊人脸.

To avoid the impact of noise in grayscale regions on the accuracy of facial recognition,this paper proposes a deep learning-based low-quality image blurred face recognition method.Firstly,we use wavelet transform to discretize low-quality images and obtain contour scale information containing different directional features.Secondly,we detect key points of facial contour lines on facial features and uniformly sample the noise by calculating the covariance matrix between the identified sample and the source sample and the membership function of the image to reduce noise content.In addition,this paper combines the threshold function of key points to match key points from multiple angles.Finally,using deep learning algorithms to construct a recognition classifier and calculating the distance between two samples in the multi-angle matching key point results we achieve face recognition.The test results show that the proposed method achieves high recognition accuracy in facial image recognition with different fuzzy ratios.