An energy-minimization approach to color constancy

A computational approach to color constancy is formulated as an energy minimization problem. The energy function for the problem consists of error terms (data fitting) of the contrast and the local DC values between input image and recovered surface reflectance, and a gray-world constraint term. A neural network model which works to minimize the energy is constructed and the performance of the model is evaluated using the images under the illuminants with color and spatial luminance variations. It turned out that the illuminant variations were removed and color constancy was achieved in the output image. It is shown that the color contrast effect can also be simulated by the proposed model.