Research Of Automatic Red-eye Removal Based On Face Detection

"Red-eye effect" is a phenomenon that when a flahgun is used, the center of pupils appears unnaturally reddish in the center of pupils. It hinders the wide application of embedded products with flashlight function such as digital camera, mobile phone and video cameras. The paper main realizes an automatic redeye removal system. The system was divided into two modules:face detection and redeye removal. The module of face detection was divided into two stages:training and detection, and the redeye removal module main consist of redeye location and redeye modification.The module of face detection adopted the adaboost algorithm. It is analyzed and realized in detail including Haar-like feature, integral image, training flow of Adaboost algorithm and the flow of detecting face. The face detection system is realized, and at last it shows effect pictures of face detection.The redeye removal module will operate after face detection. By calculate the value of redness to the face, and then implement redness segment to the redness image. The red-eye pixels are then located with several refining masks computed over the facial region. The optimal masks are created by morphological closing, then using geometric characteristics to restrict the non-redeye objects. Once the real redeye regions have been located, adjustment is made to recover the actual color of pupil region, and smooth operation is made to edge pixels of redeye region to restore natural color of pupil.At last the automatic red-eye removal system is achieved by c and c++ language on the platform of VC6.0, and a set of redeye images are collected in the experiment as test set. The method achieves a perfect rate of accuracy if the input image including a high redness eye and the frontal face detecting. But the method is existing failure effect if the input image including a low redness eye and non-frontal face which exceed more than 30 angles. However, frontal faces take the great proportion to general digital photos. So the method proposed in this paper can remove redeye in digital photos effectively and it has realistic application in digital field.