Super-resolved face images using robust optical flow

Abstract

Surveillance systems are commonly used to monitor and track individuals in a cluttered environment. Face images that are captured using such a system often suffer from poor resolution and consequently degrade the performance of any face recognition system which may be applied to these images. Super-resolution (SR) is one avenue for overcoming this limitation, however, many existing SR techniques perform poorly in applications involving the human face as faces are non-planar, non-rigid, non-lambertian, and are subject to self occlusion. This paper presents a superresolution system using robust optical flow in order to overcome these limitations. The optical flow method employed incorporates robust estimation methods to overcome problems associated with violation of the brightness constancy and spatial smoothness constraints. Resolving these issues greatly enhance the quality of the super-resolved images. Experimental results show significant improvement of the image quality and image resolution.