Study Of Skin Color Based Face Detection And Its Application In Redeye Removal

Redeye is a common phenomenon while photo is taken by flash lamp. Human's pupils will be dilated when the ambient illumination is generally low. In this condition, human's pupils can not constrict immediately when photo is taken by flash bulb, the light from the flash goes through the pupils and irradiates on microvascular organization of the retina. And then reflects back red light through the pupils, which leads to an unnatural red hue. That's the reason for redeye is generated. The color of redeye appears quite different from the color of pupil which is perceived by people. It is often regarded as an obvious defect of a photograph. A convenient and effective redeye removal method is beneficial and necessary for improving the quality of photographs.This paper proposes an automatic redeye removal method based on skin color-based face detection. This method is made up of four parts:Face detection, Eye location, Redeye detection and Redeye correction.First of all, a skin model is used for image skin-color segmentation and obtaining skin color regions of original image. After that, morphological processing and face geometrical features are worked on detecting the human face area from these skin color regions. Then, human face area is divided into small image blocks. According to eye area has the largest gray-scale complexity character in each eye possible region; this character can be used for locating the largest gray-scale complexity image block in the left or the right eye possible region which can be considered as the left or the right eye area. For locating eye area more accurately, center correction algorithm is used for adjusting the center of each eye area. In the part of Redeye detection, a red threshold associated with the Cr component in YCbCr color space is defined and a redeye mask is generated by acting this threshold on the eye area. After that, a least squares fitting method is used to find the position and location of the pupil. In the last part of Redeye correction, redeye is corrected by adjusting the R, G, B components of the red pupils' pixels. Meanwhile, pupils'edge are smoothed for obtain more natural redeye correction result.Results show the method proposed in this paper can automatically remove redeye in digital photos effectively and it is also robust and fast.