A Face Quality Evaluation Method Based on DCNN
作者单位：Beijing University of Technology
会议名称：《第32届中国控制与决策会议》
主办单位：IEEE Control Systems Society (CSS);Northeastern University;State Key Laboratory of Synthetical Automation for Process Industries;Technical Committee on Control Theory, Chinese Association of Automation
会议日期：2020年
学科分类：0820[工学-石油与天然气工程] 12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 081104[工学-模式识别与智能系统] 080203[工学-机械设计及理论] 0802[工学-机械工程] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0801[工学-力学（可授工学、理学学位）] 0823[工学-交通运输工程] 0812[工学-计算机科学与技术（可授工学、理学学位）]
基 金：National Key Research and Development Program of China NKRDPC (2017YFB030 6404)
关 键 词：Deep Learning Multi-task Network Norm Value Face Quality Evaluation
摘 要：In response to the problem of low recognition rate caused by low quality face image in face recognition, a face quality evaluation method based on deep convolutional neural network is proposed. Firstly, the depthwise separable convolution method is adopted to establish a depth-convolution model which contain 8 blocks to achieve face feature extraction. Secondly, the last layer of the convolution module connects 8 output branches, and uses regression and classification methods to predict the probability of 8 attributes, namely yaw, pitch, norm, opening mouth, occlusion, blur, dim, and closed eyes. The proposed method also proves that norm value can help analyze the quality of face image. Finally, the weight optimization is realized by taking the video pass rate as the optimal objective function. the quality score of face image can be calculated by weighting each branch The higher the score, the better the quality of the face image. In this paper, an end-to-end face quality evaluation model is implemented by using depthwise separable convolution method. The parameters of the model are less than 60000, the operation speed is fast, and the evaluation results are accurate;and it can filter out low-quality face image in real time, and recommend high-quality face image to face recognition model.