Illumination invariant face detection in the compressed domain

Abstract

Significant attention has recently been drawn to Human Robot Interaction (HRI) that uses intelligent robot to recognize the outer state and make decisions to mutually interact with human. In HRI environment, vision-based approach to identify user is known to be the most effective method. There is also a tendency of using network robot, which uses server to perform high computation processes, to relieve the computation complexity of vision-based approach. The computation load can be further lowered by using detection in the compressed domain which will also save the cost of bandwidth. However, visual information of vision based approach can be corrupted due to illumination environments. Therefore, in this paper, in order to solve illumination related problem, face detection system that works directly under the compressed domain and compensates illuminations is proposed. Experimental results verify that the proposed illumination compensation and color balancing scheme successfully eliminated illumination existed on the perceived image from the robot. Also, results shows that the proposed detection system can be usefully used in the compressed domain.