Red eye detection using color and shape

Abstract

Red eye artifact caused by flash is a common problem in consumer photography. The artifact can be avoided by certain flash techniques, but the use of image processing algorithms has proved to be an efficient approach for the detection and correction of the red eyes in the images. In this paper, we present a novel method for an automatic detection of red eyes in the image. The method is based on the localization of red regions in the image, after which the special character of each candidate region is inspected based on shape analysis. Once the candidate regions have been detected, they are classified into red eyes and other regions based on specific classification rules. In the experiments, a wide set of digital photographs containing red eyes in different conditions and circumstances have been used. The results reveal that the method is able to detect relatively accurately the red eyes in the images, and it provides robustness and fastness required in the mobile devices. Also the rate of false positive detections is low.