A novel red-eye removal approach in digital color photos

Abstract

Red-eye artifact is induced by the light reflected off retinal capillaries when photos are taken under weak light using flash. An automatic method is proposed to remove red-eye artifacts in digital images in this paper. First, a face detector based on histograms of oriented gradients (HOG) features and Real Adaboost algorithm is employed to select faces, and eye-boxes which contain one eye in each of them are obtained from the detected faces. Next, a mask which records candidates for red-eye is calculated by two segmentation criteria computing over each eye-box. Afterwards, the candidate mask is processed by morphological filters and several heuristic exclusive principles to locate the target red-eye region. At last, the red-eye defect is compensated by an adaptive correction technique. The proposed method is efficient and effective when tested on real images taken by different digital cameras.