Binarization of degraded document image using Gaussian Markov random field model

This paper presents a binarization approach to degraded document images, which is based on Gaussian Markov Random Field (GMRF) model. The energy function with the single-site and pair-site clique potential functions is formulated for the GMRF. The parameters of the potential functions are estimated by expectation-maximization (EM) algorithm, without necessity of training process. Experiments on different types of degraded document images with various noise, contrast variation or uneven illumination, have demonstrated the validity of the proposed method.