Face Identification Under Disguise and Makeup Based on Hybrid Deep Learning

Abstract

Facial recognition has been broadly used in advanced intelligent systems (i.e: smart video surveillance, intelligent access control system, and online payment). The performance of existing algorithms for automatic facial recognition is hampered by various covariates like pose variations, face aging, disguises, and makeup. Disguises and makeup are especially used to intentional or unintentional changes facial appearance to either hide one's personal identity or impersonate someone's different identity. While new algorithms continue to improve performance, most face recognition systems are liable to failure when disguised or makeup altered, which is one of the most challenging factors to overcome. With enormous capability and promising results, deep learning technology becomes attracted to the greatest attention to the research in a diversity of computer vision tasks. In order to overcome this problem, the database of disguised and makeup faces (DMFD) is used. In this paper, face features are extracted by Linear Discriminant Analysis (LDA). Facial recognition is done by using proposed hybrid-deep learning Classifier for more precise feature learning. Also, we compared the proposed method with two pre-trained models (AlexNet and VGG16). The Experimental results taking after implementation and testing showed the effectiveness of the proposed system provided better precision by (94%)