Deep Learning based Face Recognition System using Dual Shot Face Detector and Face Landmarks

Abstract

Background/Objectives: Face detection and identification system have become an essential system in the image-security-system area. The present study aims to develop a robust face detection and identification system in complex scenes using Deep Learning, Dual Shot Face Detection (DSFD) algorithm, and face landmarks. Methods/Statistical analysis: The face detection process is designed by the DSFD algorithm, which has the best accuracy rate in the face detection area recently. In addition, we use the WIDER Face Dataset at the proposed system network training, which is the famous face detection training and testing DB. In addition, the identification process is composed of the face landmarks vector information using deep learning network.
Findings: Recently, state-of-the-art face detectors can be roughly classified into two-stage (R-CNN) detection and one-stage (SSD, YOLO) methods. However, one-stage face detection architecture has fascinated more attention due to its higher inference ef?ciency and fast system deployment. In the experiment, we can find that the detection accuracy depends on the face detection algorithms applied to it. Improvements/Applications: In the experiment result, we could find that the proposed method showed better face recognition performance compared to the conventional SSD based face recognition method.