Automatic Redeye Correction Algorithm with Multilevel Eye Confirmation

Abstract

Redeye is a well-known problem in portrait photography. This effect is caused by the light entering the subject's eye through the pupil and reflection from the retina back to the imaging sensor. Many of the existing redeye correction methods sometimes fail to detect the actual redeye artifacts and incorrectly change other red colored areas. The proposed algorithm describes a fully automatic redeye correction system with multilevel eye confirmation stages. The algorithm first identifies the skin and locates the redeye region using color information. The detected region is then confirmed as redeye by its redness variation, glint, eye-lips triangle, and comparison with surrounding regions. The algorithm removes the falsely extracted components by verifying with rules derived from the spatial and geometrical relationships of facial components. Finally the defect is corrected by desaturating the red pixels while preserving the natural glint presence.