In a document image enhancement method, text character strokes are enhanced in a way such that areas closer to the center (skeleton) of each stroke are enhanced (e.g. made darker) by greater amounts than areas farther away from the center. Each text or line region of the input image is first binarized to generate a binary image containing connected components corresponding to character strokes and lines. Multiple levels of pseudo-skeletons are computed for each connected component, for example by using successive thinning. Multiple pseudo-skeleton difference areas, i.e. differences between successive levels of pseudo-skeletons, are generated. Pixels located in different pseudo-skeleton difference areas are enhanced by different amounts, by applying different inverse-degradation functions. Graphical regions of the input image may be treated with edge enhancement.