A new combining learning method for color constancy

The ability to measure color of objects, independent of color of the light source, is called color constancy which is an important problem in machine vision and image processing. In this paper, we propose a method that employs a neural network to estimate the chromaticity of light source. This network uses the results of four well known color constancy methods as its input in training and tries to find the best result in test phase. In selecting the input methods, it has been tried to select ones which each one focuses on a particular specification of the colored image and is suitable for training also. By considering these issues, Max RGB, gray world assumption, gray edge, and shades of gray as well known methods were selected. In the proposed methods, the result in test phase may correspond with none of these algorithms necessarily. The experimental results showed that the proposed method reached to a good estimation of the illuminant source with less complexity in comparison to the previous related works.