Multi-illuminant Estimation Based on Spatial Fuzzy C-means Clustering with Application to Image White Balance

Abstract

In most scenes, images captured by digital devices are greatly influenced by environmental illuminant. Because there are different illuminants in the scene, the color from objects even change and be similar to the illuminant color. It results that the objects cannot reflect the true color and this phenomenon is called color cast. In this thesis, by utilizing basic color science and experimental results, a dataset of color sample under different illuminants is constructed. Using the existing method of illuminant estimation to calculate plural color temperatures of the input image, called the initial estimated color temperature. Furthermore, utilizing clustering algorithm to execute illuminant classification to initial plural color temperatures, and calculating the multi-determined color temperatures of image. This study takes the lead in utilizing the spatial fuzzy c-means (SFCM) clustering algorithm to execute the illuminant classification. In traditional white balance, the correction configuration focuses on single illuminant scene, which is fixed correction value, whereas there are multi-illuminant scenes mostly in real scene, the performance of white balance is not good. In this thesis, the adopted SFCM algorithm revises the main illuminant individually and solves the color cast successfully.