Estimation of illuminant chromaticity based on highlight detection for face images with varying illumination

Illuminant chromaticity estimation is required in the fields of the tracking and recognition of human faces, compensation and detection of skin color, image reproduction, and color constancy. This article proposes a simple and effective method to estimate the illuminant chromaticity using highlights from the face region. We first select the candidate points through the detection of highlights around the leftmost pixel of the rg chromaticity space from the facial images under various illuminant conditions and then estimate the illuminant chromaticity by averaging the chromaticities of the candidate points. To enhance the illuminant chromaticity estimation accuracy, we propose a more elaborate estimation method, which effectively readjusts the candidate points of the highlight regions according to the conditions derived from two ethnic and three illuminant groups. The experiment results show that our proposed method reliably estimates the illuminant chromaticity under various color temperatures and illuminant directions. © 2012 Wiley Periodicals, Inc. Col Res Appl, 39, 188–199, 2014