A Novel Red-Eye Correction Method Based on AdaBoost Algorithm for Mobile Phone and Digital Camera

Abstract

Caused by light reflected off the subject's retina, red-eye is a troublesome problem in consumer photography. Correction of red eyes without any human intervention is an important task. There are some algorithms existing for red-eye detection, but almost all of them have less accuracy, in addition, they cannot support both high pixel and single red-eye. In this paper, a novel approach is proposed to eliminate red-eyes in the digital images automatically with a satisfactory result. This method gets the face region first by AdaBoost algorithm and then detects the red-eye on the top part of the face region, corrects the red-eye in the eye region for recovering the image's original color at last. Experiments in the platform of mobile phone and digital camera show that this method can eliminate the red-eye with high accuracy of 87%, which is higher than the best known technology of face detection base on complexion by 7%, and it can also support 8 million pixels image, moreover, the method has advantages of robustness and real-time computability.

Development of a geometrical algorithm for eye detection in color images

“…Eye detection is mostly found in many applications such as criminal investigation, drowsy driver detection, red eye correction and blink detection in a digital camera [1][2][3][4]. Many machine learning algorithms such as AdaBoost [5], Artificial neural network [6] and Support vector machine algorithm [7] are normally used in the eye detection system.…”