Skip to main content

Recognition of Challenging Ocular Images

Researcher: Vijayakumar Bhagavatula

Research Area: Security of Cyber-Physical Systems

Abstract

Scope: Iris recognition is known to offer very good human identification performance among all biometric modalities --- but this is true only under highly controlled scenarios where the acquired iris images are of high quality. In real-world applications, the image quality degrades due to several factors, e.g., the subject being far away from the camera (i.e., low resolution), partially closed eyes, non-frontal gaze, specular reflections, etc. The goal of this project is to investigate whether using ocular region (i.e., iris region plus the area near the iris such as eyebrows, skin texture, eye corners, etc.) can be used to achieve improved identification performance.

Outcomes: As a performer in the Intelligence Advanced Research Project Activity (IARPA)’s BEST program, we developed several matching methods that can match challenging ocular images. On the challenging Face and Ocular Challenge Series (FOCS) ocular image data set, we achieved an equal error rate (EER) of about 18% using ocular regions whereas iris-matching algorithms yield an EER of about 34%. Current research is aimed at further improving the performance on challenging ocular images.