A Novel Color Correction Framework for Facial Images

Abstract

The color images produced by digital cameras are usually not in conformity with their inherent colors. This will seriously impact computer-aided facial image analysis because it is on the basis of accurate rendering of color information. To solve that, we propose a novel color correction framework. Firstly, we utilize 122 undistorted facial images to demarcate complexion gamut. Secondly, several training sets based on complexion gamut are compared experimentally for the selection of optimal training samples. Thirdly, we select an adaptive target device-independent color space for our facial images color correction task. Finally, we evaluate the performance of three most popular color correction algorithms in color science area, and select the most suitable one to build our final regression model. Compared with the previous work, our color correction framework is characterized by mission dependence and statistical reliability. Besides, its trained model has low complexity and high accuracy. All of these features make it effective for facial images color correction.