Automatic white balance based on estimation of light source using fuzzy neural network

In this paper, we propose an image processing pipeline to estimate the color temperature and white balance an image. The method comprises a pre-processing stage to generate training samples for the fuzzy neural network (FNN), which plays the role of predicting pixel-level color temperatures, and two post-processing stages to integrate the outputs of the FNN into an overall color temperature of and white balance the image. Experimental results and comparison study verify that the proposed method performs better than most existing methods.