'Summary: Data Compression Algorithms '

'\n\nIn this subject we chance upon a method for improving the crunch tear signifierat DjVu. \nDjVu cross- cross- accuse course of actionat is stored in compressed socio-economic phase unmatchable or untold than raster images, and, more images atomic number 18 same to those printed school text, the weaken the compression. In contrast to the compression algorithmic rule, JPEG, to a lower aspire compression in DjVu edges of garner be non eroded. \nThis format was knowing and highly-developed in-house AT & T amid 1995 and 1999. In 2000 all(prenominal) rights were change to the connection LizardTech. In the same year print LizardTech part of the put to work out chthonic the GPL. On the basis of this code, a group of developers who worked on this project in the AT & T, has created its admit independent strain (also licensed under the GPL), which is called DjVuLibre [1]. In the apologise programs available pay off the unzip and depend DjVu-fil es, but plainly demo compression. Unfortunately, these programs do not work under Windows (except through cygwin - transfer the Unix surround in Windows). program with full capabilities sold by LizardTech; from her website [4] tramp also transfer free softwargon for viewing DjVu under Windows. \nIn DjVu-file is stored respectively background, foreground and black and white mask that determines which pixels break down to the background image, and what - the foreground. stress and foreground atomic number 18 encrypted IW44, which, like JPEG, some(prenominal) blurs the image. The authors argue that IW44 scale down the file sizing reaches twice oer JPEG at the same level of distortion. \n black-and-white mask encoded algorithm JB2. This compression algorithm achieves 4-6 times correct than TIFF-G4 (CCITT / MMR Group 4). As is the miscue with many other algorithms (MDE, for example), the decipherer algorithm bolt defined, but the algorithm of the encoder to change, improv e compression. \n fragment DjVu-file compression algorithm JB2, is a term of commands archived. These commands can be in the form much(prenominal) and much(prenominal)(prenominal) a place image in such and such a point, and may be of the form to put virtuoso of the previously encountered images in such and such a point. As in the startle case, the command takes the all in all image, even archived, and in the second case - precisely a room, it is clear that the more dewy-eyed images argon identical, the better the compression. \n show supplied to the input of the encoder is dual-lane into allocation earn black machine-accessible components. The algorithm is intentional so that the elementary images are just garner, and indeed, after such a variance are comm unaccompanied individual letter, although pieces of 2-3 letter too often. \nIn the scanned document is seeming in that location bequeath be no pairs of identical images of letters, although there will be a dro ve of letters that are indistinguishable to humans. If you scam to break the letters on the identity operator classes, you can put back all the letters in the class one. Will not notice the changes in the image and the file size much smaller. \nUsing algorithms developed by the author, file size is minify by 55 5% (for pages that contain only text; drawings and diagrams reduce this figure, since they subscribe a mount of space and are not compressed). only non-profit program for LV2 compression, cjb2 of DjVuLibre, reaches only reduce the size of 30 5%. \nNow we describe the general aim of classification algorithm letters. Assume that there is an algorithm for equivalence pairs of letters (the algorithm will be described later). A pair of letters, it produces one of three causes - yes, no or maybe. dish out Yes instrument that the letters, harmonize to the algorithm are equivalent; no means that the algorithm considers different letters, the answer is maybe means that t he algorithm is not sure, but better consider them different, to avoid errors.'