A deep learning framework for the computational color constancy problem is proposed.

Illumination is estimated by solving an illumination classification problem.

Convolutional neural network is trained to solve the classification problem.

The proposed method outperforms previous methods on multiple test datasets.