On compression of multi-level document images

Most original document images tend to be bi-level. When electronically captured, raw pixels are represented typically as a 6 or 8 bit number on a rectangular array. Traditionally, these pixels are thresholded to generate a binary image for further processing. In the first part of the paper we describe why is it advantageous to threshold an image to a 2- or 3-bit pixel image rather than a 1-bit per pixel image. This helps mitigate the staircasing effect (seen in a binarized image), especially when the spatial resolution of the sampled image is poor. The second part of the paper describes several encoding algorithms to compress multi-level thresholded images. Through examples, it is shown that compression of error images (generated by DPCM), utilizing a modified runlength coding than the CCITT Group 3 coding and when pixels are concatenated, yields substantially better compression factors than the other approaches.