Color Constancy Using Ridge Regression

Although there exist a number of single color constancy algorithms, none of them can be considered universal. Consequently, how to select and combine existing single algorithms are two important research directions in the field of color constancy. In this paper we use ridge regression, a simple yet effective machine learning approach, to select and combine existing color constancy algorithms. Two algorithms are proposed using ridge regression in this paper. In the first method (combination of existing color constancy algorithms), the color of the light source is estimated based on the results of existing single color constancy algorithms using ridge regression. In the second method (selection of existing color constancy algorithms), the proper algorithm for a given image is selected by ridge regression based on natural image statistics. Then it is used to estimate the color of the light source. The proposed two algorithms are verified on a large data set with more than 11,000 images. The experimental results demonstrate that the proposed methods are competitive with the state-of-the-art methods.