New segmentation techniques for document image analysis

摘要

In this paper, some raster image segmentation techniques are proposed to extract character/graphic pixels from scanned document images. These methods are very efficient in extracting character/graphic pixels from the images with a variety of background types. The performance of these methods is compared with Lloyd's1 method, the logical level technique2, Otsu's3 method, dynamic thresholding4, the nonlinear adaptive techniques and the integrated function technique5. Experiments are carried out with a number of scanned document images and artificially generated text images.