We present a fully automatic approach to the detection and correction of red-eye artifacts, as found in consumer photographs. The approach is model-based in that we first detect faces, then detect eyes in the found faces, and finally determine whether the eyes show any red-eye defects. Provided the detection stage found some red-eyes, we attempt to replace the red color by a suitable replacement color. Especial care is taken to ensure smooth transitions between corrected and original parts of the eyes.
Andreas Held, "Model-Based Correction of Red Eye Defects" in Proc. IS&T 10th Color and Imaging Conf., 2002, pp 223 - 228, https://doi.org/10.2352/CIC.2002.10.1.art00042