home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.compression.research
- Path: sparky!uunet!mcsun!sunic!kth.se!hemul.nada.kth.se!md85-epi
- From: md85-epi@hemul.nada.kth.se (Urban Koistinen)
- Subject: Re: arithmetic coding
- Message-ID: <1992Nov16.112726.4493@kth.se>
- Keywords: when does it work ?
- Sender: usenet@kth.se (Usenet)
- Nntp-Posting-Host: hemul.nada.kth.se
- Organization: Royal Institute of Technology, Stockholm, Sweden
- Date: Mon, 16 Nov 1992 11:27:26 GMT
- Lines: 27
-
- In article <Nov.15.22.54.42.1992.14006@pepper.rutgers.edu> phalke@pepper.rutgers.edu (V. Phalke) writes:
- Supposing SIGMA = { 0, 1}
- If the probability P(0) is very high and P(1) is close to ZERO arithmetic
- coding doesn't seem to work !!
- I mean supposing the string is pure ZEROES then arithmetic coding interval
- will decrease very slowly. This would result in a very long encoding. On
- the other hand the entropy of this data is very low!!
- Any help ??
-
- If the interval doesn't decrease much, few bits need to be sent
- giving good compression.
- Many bits in, few bits out.
-
- phalke@paul.rutgers.edu (908) 878-1602(h) (908) 932-4635(o)
-
- The great thing about arithmetic coding is that if you know how to
- calculate the probabilities for the next character, you get a code
- that is very near optimum.
- The bandwith wasted by the encoding scheme is only about exp(-N*log(2))
- per bit encoded.
- (N is the number of bits used by the coder)
-
- This mean, the quality of compression depend only on the statistical
- model.
-
- --
- Urban Koistinen - md85-epi@nada.kth.se
-