Safe Red-Eye Correction Plug-in Using Adaptive Methods

Abstract

An important issue with red-eye correction is that it might result in image degradation. This can be due to the detection of false positives or, even in the case of correct detection, to an inadequate correction technique. Three correction methods are proposed and compared according to their image degradation risk and their expected perceptual quality improvement. Based on those analyses an adaptive system is designed which selects the correction strategy dependent on those measures and the detection confidence. Finally, both qualitative (visual preferences) and quantitative (pixel counts on manual segmented images) evaluation results are shown.