Color constancy is an image processing problem to correct for colors in digital images distorted by noise and illuminants. There are several scientific applications in which the true color of an object in a digital image is needed for analysis. Without calibrating the colors in the image, the recorded pixel values in the image do not accurately define the true colors. In this thesis, we propose two innovative methods: an empirical Bayesian method and a Color Convolution method to calibrate colors in images. We utilize 20 RAW images from a color constancy image database and compare our methods against the Grey-World assumption, White Patch assumption, and existing Bayesian techniques. Standard convergence techniques verify the stability of Monte Carlo Markov Chain in our Bayesian method and adding regularization to the Color Convolution method provides numerical stability. Our results show that the proposed algorithms outperform other existing approaches by the mean square error of the pixel values between images.
展开▼
