Blurred Face Image Authentication for Enterprise Attendance Using Adaptive Light Adjustment and GAN-CNN Architecture

Abstract

This paper combined the adaptive light adjustment algorithm and the generative adversarial network (GAN) deblurring algorithm with a convolutional neural network (CNN) algorithm for blurred face image recognition. First, the adaptive light adjustment algorithm and the GAN algorithm were used to perform deblurring operations on blurred face images, and then the CNN algorithm was used to recognize them. Then, simulation experiments were conducted. In the experiments, the adaptive light adjustment-combined GAN deblurring algorithm was compared with the Gaussian filter method and traditional GAN algorithm. The proposed face authentication algorithm was compared with the support vector machine and traditional CNN algorithms. The results showed that the adaptive light adjustment-combined GAN algorithm could effectively deblur the face image, with a peak signal to noise ratio of 32.08 and a deblurring time of 0.29 s. Moreover, the proposed face authentication algorithm could effectively recognize the identity of the blurred face image, with a precision of 0.987, a recall rate of 0.986, and an F value of 0.986, and it consumed 0.31 s for recognition.