A Deep Learning Paradigm for Automated Face Attendance

Abstract

In this paper, we propose an end-to-end automatic face attendance system using Convolutional Neural Networks (CNNs). Attendance of a student plays an important role in any academic organization. Manual attendance system is very time consuming and tedious. On the other hand, automatic attendance system through face recognition using CCTV camera can be fast and can reduce the man-power involved in that process. Here, we have pipelined one of the best existing architectures such as: (i) Single Image Super-Resolution Network (SRNet) for image super-resolution, (ii) MTCNN for face detection and (iii) FaceNet for face recognition in order to come up with a novel idea of marking attendance. Due to poor video quality of CCTV camera, it becomes difficult to detect and recognize faces accurately and this may reduce the attendance accuracy. To overcome this limitation, we propose a CNN framework called SRNet which super-resolves a given low resolution (LR) image and also increases the face recognition accuracy. We make use of five different datasets i.e. RAISE and DIV2K for SRNet, VGGface2 for FaceNet, LFW and our own dataset for testing and validation purpose. The proposed face attendance system displays a sheet which consists of a list of absent and present persons and the overall attendance record. Our experimental results show that the proposed approach outperforms other existing face attendance approaches.