Text extraction from degraded document image independent of character color based on MAP-MRF approach
Hideaki Kawano, Hideaki Orii, Hiroshi Maeda, Norikazu Ikoma · 2009
We propose a novel background and foreground estimation algorithm in MAP-MRF approach for binarization of degraded document image. In the proposed algorithm, an assumption that background whiteness and foreground blackness is not employed differently from the conventional algorithm, and we employ characters' irregularities based on local statistics. This makes the method possible to apply to the image with various colored characters, ex. outlined characters by colored background. Experimental results show that the proposed method achieves promising results.