Segmentation of document images

Several methods for segmentation of document images are explored. The authors pose the segmentation operation as a statistical classification task with two pattern classes: print and background. A number of classification strategies are available. All require some prior information about the distribution of gray levels for the two classes. Learning (either supervised or unsupervised) and automatic updating of the class-conditional densities are performed within image subregions to adapt global density estimates to the local area. After local densities have been obtained, each pixel within the window is classified; several techniques for this are considered. Results on four test images indicate that the commonly used contextual models are not suitable to all document images. >