In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
will defend his proposal
Illumination-Invariant Face Recognition
AbstractRecent advances in face recognition technology have created many useful applications for humans, such as security enhancements, mobile face unlock, and smart doorbells. However, facial recognition algorithms remain challenges when identifying images with many changes in appearance due to camera angles, lighting conditions, and occlusions. Poor lighting conditions are one of the biggest challenges for face recognition algorithms. The performance of state-of-the-art, face-recognition algorithms drastically drops when measured on datasets with significant illumination variations. Thus, the goal of this dissertation is to achieve statistically significant improvements to the performance of face-recognition systems using 2D images that depict individuals exhibiting changes in illumination. To achieve this goal, three primary objects are proposed, namely (i) collect and annotate facial data from indoor and outdoor environments, (ii) develop and evaluate an algorithm for matching face images to overcome appearance changes due to illumination variation, and (iii) develop and evaluate an algorithm for matching face images to overcome perception changes due to extreme illumination conditions. In this proposal, two datasets are collected and annotated to facilitate the evaluation of face-recognition algorithms under illumination variations. A series of facial reconstruction and relighting methods are proposed and employed as a data augmentation method to enrich face datasets with illumination variations. Also, a semi-supervised low-light face-enhancement method is introduced to enhance the lighting conditions of low-light face images.
Date: Friday, May 24, 2019
Time: 11:00 AM - 12:30 PM
Place: HBS, Room 317
Advisor: Dr. Ioannis A. Kakadiaris
Faculty, students, and the general public are invited.