Lossless Compression of Digital Images

Presently, tree coders are the best bi-level image coders. The currentISO standard, JBIG, is a good example.By organising code length calculations properly a vast number of possible models (trees) can be investigated within reasonable time prior to generating code.A number of general-purpose coders are constructed by this principle. A multi-pass free tree coding scheme produces excellent compression results for all test images. A multi-pass fast free template coding scheme produces much better results than JBIG for difficult images, such as halftonings. Rissanen's algorithm `Context' is presented in a new version that is substantially faster than its precursorsand brings it close to the multi-pass coders in compression performance.Handprinted characters are of unequal complexity; recent work by Singer and Tishby demonstrates that utilizing the physiological process of writing one can synthesize cursive writing in a mostconvincing manner. The models which are generated for different letters display not only different parameter values as would be expected but highly varying model orders. A common description ofall alphabet symbols seems therefore unobtainable. However, letters which are confusedby human beings and by man-made OCR systems usually have approximately the same appearence andmay therefore be modeled jointly. We part the set of bitmaps into types, where each type has itsunique feature space.The bitmaps belonging to some type is modeled independently from bitmaps belonging to other types.The feature vector of a bitmap initially constitutes a lossy representation of the contour(s) of the bitmap. The initial feature space is usually too large but can be reduced automatically by use ofa predictive code length or predictive error criterion.