Bayesian Colorization Using MRF Color Image Modeling

This paper presents a colorization algorithm which produces color images from given monochrome images. Unlike previously proposed colorization methods, this paper formulates the colorization problem as the maximum a posteriori (MAP) estimation of a color image given a monochrome image. Markov random field (MRF) is used for modeling a color image which is utilized as a priori information for the MAP estimation. Under the mean field approximation, The MAP estimation problem for a whole image can be decomposed into local MAP estimation problems for each pixel. The local MAP estimation is described as a simple quadratic programming problem with constraints. Using 0.6% of whole pixels as references, the proposed method produced pretty high quality color images with 25.7 dB to 32.6 dB PSNR values for four standard images.