STUDYING OF DUAL DEEP PERIOCULAR PARTS FOR PERSONS RECOGNITION

Abstract

Recently, periocular region has been employed in recognitions and it can be so effective especially in wearing a face mask as happened during the Coronavirus pandemic. In this study, a new method is proposed for recognizing persons based on their perioculars. It is named the Dual Deep Periocular Parts (DDPP). In this method, two deep learning networks are employed, where each network is determined for a certain periocular side (right or left). They are termed the Deep Network for the Right Periocular (DNRP) and Deep Network for the Left Periocular (DNLP). Both the DNRP and DNLP are fused together to construct the proposed DDPP approach. Also in this paper, a database called the Northern Technical University Periocular Database (NTUPD) is collected from scratch. Persons recognition based on the proposed periocular approach shows further performance enhancements as we obtained results of accuracy that reached 98.7% and Equal Error Rate (EER) 1.3%.