Document image binarization using a discriminative structural classifier

摘要: A new supervised algorithm for document image binarization is proposed.The proposed method uses a discriminative graphical model binarization.The results are compared with the participants of famous contest. Binarization one key initial steps in analysis and system understanding. Different types degradations make very challenging task. This paper proposes statistical framework binarizing degraded images based on concept conditional random fields (CRFs). The CRFs models which distribution used structural classifications. binarized given ones modelled respect to set informative features extracted all sites image. recent marginal learning 5 estimation parameters model. enables depending labelling despite independent pixel-by-pixel other methods. performance our evaluated different datasets several well-known Experimental show comparable state-of-the-art