<title>Lossless/lossy compression of bilevel images</title>

Abstract

We present a general method for lossless/lossy coding of bi- level images. The compression and decompression method is analogous to JBIG, a current international standard for bi- level image compression, and is based on arithmetic coding and a template to determine the coding state. Loss is introduced in a pre-process on the encoding side by flipping pixels in a controlled manner. The method is primarily aimed at halftoned images as a supplement to the specialized soft pattern matching techniques which work better for text. The new algorithm also works well on documents of mixed contents e.g. halftoning and text without any segmentation of the image. The decoding is usually slower than JBIG due to a more wide-spread template. A decoding output of more than 1 Mpixels per second can be obtained in software implementations. We present a greedy 'rate-distortion' algorithm for flipping as well as less complex algorithms intended for relatively fast encoding and moderate latency. In the less complex algorithms, flipping and encoding is carried out in the same pass. The potential risk of flipping avalanches is minimized by conditioning flipping on the sign and magnitude of the local gray-scale error computed by a forgetful error diffusion algorithm. Template based refinement coding is applied for a lossy-to-lossless refinement step. The (de)coding method is proposed as part of JBIG-2, an emerging international standard for lossless/lossy compression of bi-level images.