Color Constancy Based on Deep Residual Learning

The purpose of color constancy algorithm is to eliminate the influence of illumination on the color of objects in the scene, so that the computer has the same color constancy ability as human visual system. In order to further improve the accuracy and robustness of the color constancy algorithm, this paper proposes a illumination estimation method based on deep residual learning, which fully extracts the illumination feature information in the image by deepening the number of network layers, and uses the residual module to prevent over fitting of the network model, At the same time, the local illumination estimates are integrated to obtain the global illumination estimation of the whole image. The experimental results on ColorChecker data set show that the estimation accuracy and robustness of this method are good, and can be applied to the fields of image processing and computer vision requiring color correction.