To Perform White Balance Using Skin Memory Color

ABSTRACT In the imaging industry, color reproduction is a critical aspect, and white balance plays a pivotal role in ensuring color reproduction. In this research, an automatic white balance algorithm based on the known color center from a color target, such as the Preferred Memory Color Chart was developed to achieve preferred color reproduction. First, a target detection algorithm YOLOv5 based on CNN was implemented to identify the face region in the image. The skin color ellipsoid was then used to create a skin mask to perform illumination estimation based on the skin color. Finally, the chromatic adaptation transform, the CIECAT16 model, was used to render the image to D65. Additionally, an image dataset was accumulated to evaluate the effectiveness of the algorithm. During the testing process, it was observed that the different colors of the background in images had a significant impact on the light source estimation. In response to this, a simple linear model was developed to mitigate the interference caused by background colors to some extent.