From ajr@eng.cam.ac.uk Thu Aug 11 00:11:23 1994 Path: saips.cv.nrao.edu!hearst.acc.Virginia.EDU!concert!gatech!swrinde!pipex!lyra.csx.cam.ac.uk!nntp-serv.cam.ac.uk!ajr From: ajr@eng.cam.ac.uk (Tony Robinson) Newsgroups: comp.compression Subject: Re: Rice Algorithm Date: 08 Aug 1994 07:22:29 GMT Organization: Engineering Department, Cambridge University, England. Lines: 59 Message-ID: References: <320dbq$35i@ucthpx.uct.ac.za> NNTP-Posting-Host: cyan.eng.cam.ac.uk In-reply-to: brendt@dip1.ee.uct.ac.za's message of 6 Aug 1994 16:22:18 GMT In article <320dbq$35i@ucthpx.uct.ac.za> brendt@dip1.ee.uct.ac.za (Brendt Wohlberg) writes: > > Does anyone have any information/references relating to the Rice > Algorithm for lossless image compression? Yes - it is the simplest Huffmann coding that you can imagine. The nth code is n bits long, hence has a probabilty of 2^{-n}. For simplicity, start with a run of zeros and add a 1 terminator: code value 1 0 01 1 001 2 0001 3 00001 4 000001 5 It can easily be extended to cope with any signal variance. For example, add another two bits at the bottom end: code value 100 0 101 1 110 2 111 3 0100 4 0111 7 00100 8 The optimal number of final bits to be added is: n = \log_2 \left( \log(2) {\sigma \over \sqrt 2 } \right) = \log_2 \left(\log(2) E(|x|) \right) I think the proper reference to this work is: @techreport{Rice91, author= "Robert F. Rice", title= "Some Practical Noiseless Coding Techniques, {Part II, Module PSI14,K+}", type= "JPL Publication", number= "91-3", institution= "Jet Propulsion Laboratories", month= nov, year= 1991} It works suprisingly well. If you want to test your code out I can provide a working implementation in: svr-ftp.eng.cam.ac.uk:/comp.speech/sources/shorten-1.14.tar.Z But this is designed to work on N independent data streams (i.e. stereo audio), not images. Tony Robinson From bhaskara@fc.hp.com Thu Aug 11 00:11:57 1994 Newsgroups: comp.compression Path: saips.cv.nrao.edu!hearst.acc.Virginia.EDU!concert!gatech!howland.reston.ans.net!vixen.cso.uiuc.edu!sdd.hp.com!hplabs!hplntx!bhaskara From: bhaskara@fc.hp.com (Vasudev Bhaskara) Subject: Re: IDCT algorithms Sender: news@hpl.hp.com (HPLabs Usenet Login) Message-ID: Date: Wed, 10 Aug 1994 23:52:54 GMT References: <324nd2$okc@styx.uwa.edu.au> Nntp-Posting-Host: hplvab.hpl.hp.com Organization: Hewlett-Packard Laboratories, Palo Alto, CA X-Newsreader: TIN [version 1.2 PL2] Lines: 31 Michael Simmons - faculties (michael@ecel.uwa.edu.au) wrote: : I have been doing some reading up on IDCT algorithms : Am I correct in saying that the algorithm presented in the paper, : "A Forward-Mapping Realization of the Inverse Discrete Cosine Transform" : by Leonard McMillan and Lee Westover of SUN, : is the best for Microprocessor based MPEG decoding. : I have read about 5 papers now and as far as I can tell they : are all pretty similar. : Is there any particular paper that any one could recommend to me. : The McMillan-Westover method uses a cache (essentially a lookup table) and depending on your processor this may or may not be fast. In fact, for faster decoding, the paper "Statistical IDCT for Image Compression", Andy C. Hung and Teresa H.Y. Meng in SPIE, Vol.2187, Feb. 1994 shows the decoding speed (in kilo blocks/sec) for the Westover method versus others and depending on compression ratio and processor, the forward mapping scheme was not that fast. for example, on the Lena image (for JPEG; the same should hold for MPEG data too), on a Sparc, faster methods yield a > 2X speedup over the forward mapping scheme and the same seems to hold for implementation on a MIPS machine. the faster scheme is an extension of the forward mapping approach wherein some symmetries are exploited in the IDCT kernel. We have a faster scheme which gives realtime MPEG1 decompression on the HP workstations but the code is not in the public domain. - Vasudev B.