Biometric recognition of surgically altered periocular region: A comprehensive study

Abstract

Wide acceptance of biometrics as an authentication mode has led to investigation of multiple modalities such as face, periocular, iris for the long term robustness. Due to various deformities arising out of deteriorating health, need for enhancing the beauty by choice or to fix the injury as a result of trauma or aging, people tend to undergo surgery. However, such surgeries do not guarantee the restoration of physical biometric characteristics (face, periocular, iris etc.) to original appearance and thereby impacting the performance in biometric identifications. Among many physical biometric characteristics, periocular recognition is widely accepted for authentication purposes. This work studies the impact of periocular surgeries on biometric performance. To this extent, we introduce a new large scale periocular surgery database comprising of 402 unique periocular images acquired before and after the surgery. This is the first work that provides comprehensive study on evaluating the impact of surgeries in periocular region on periocular recognition. Extensive experiments are carried out on a newly created dataset using 11 different state-of-art periocular recognition schemes. Further, we also explore score level fusion of these algorithms. Results obtained on the newly created large scale database indicate the degraded identification performance of both the state-of-art and fusion algorithms.