Surveillance systems are commonly used to monitor andudtrack individuals in a cluttered environment. Face imagesudthat are captured using such a system often suffer from poorudresolution and consequently degrade the performance ofudany face recognition system which may be applied to theseudimages. Super-resolution (SR) is one avenue for overcomingudthis limitation, however, many existing SR techniquesudperform poorly in applications involving the human faceudas faces are non-planar, non-rigid, non-lambertian, andudare subject to self occlusion. This paper presents a superresolutionudsystem using robust optical flow in order to overcomeudthese limitations. The optical flow method employedudincorporates robust estimation methods to overcome problemsudassociated with violation of the brightness constancyudand spatial smoothness constraints. Resolving these issuesudgreatly enhance the quality of the super-resolved images.udExperimental results show significant improvement of theudimage quality and image resolution.
展开▼
