Xref: utzoo comp.compression:736 alt.sex.pictures.d:4365
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!ub!galileo.cc.rochester.edu!rochester!pt.cs.cmu.edu!g.gp.cs.cmu.edu!tgl
From: tgl@g.gp.cs.cmu.edu (Tom Lane)
Newsgroups: comp.compression,alt.sex.pictures.d
Subject: Re: JPEG compression errors ?!
Message-ID: <13405@pt.cs.cmu.edu>
Date: 11 Jun 91 23:53:15 GMT
References: <1991Jun4.223719.2958@qualcomm.com> <1991Jun11.144958.4839@uicbert.eecs.uic.edu>
Organization: Carnegie-Mellon University, CS/RI
Lines: 37

In article <1991Jun11.144958.4839@uicbert.eecs.uic.edu>, eddins@uicbert.eecs.uic.edu (Steve Eddins) writes:
> [concerning the JPEG lossless mode]
> This part of the standard does *not* use the DCT and
> incurs no round-off errors.  According to the draft, you can expect
> 2:1 compression with the lossless coder for "typical" scenes (whatever
> that means).

That means 2:1 compared to an uncompressed full-color image (24 bits/pixel).

To put this in perspective, here's some actual data for a small (205 x 250)
full color image that I have handy.  This would be about a factor of 4
smaller than a typical 640x480 image.

	Format					# Bytes		Compression
    full color uncompressed (PPM format)	150 Kb
    same passed thru Unix "compress"			(enlarges file 9%)
    alleged performance of JPEG lossless spec	 75 Kb		2:1
    256-color GIF (made with PPM tools)		 49 Kb		3:1
    Lossy JPEG at "typical" quality setting	  9.5 Kb	16:1
    Lossy JPEG at "high" quality setting	 16.5 Kb	9:1

(The JPEG lossless number is a guess based on the 2:1 ratio claimed in the
spec; all the rest are actual numbers.  Many GIFs are more like 4:1 or 5:1
smaller than full color equivalents, so this test case might be atypical.)

Note that all but the GIF image are full color images; in my opinion the
GIF image loses a good deal more, relative to the original, than either
of the lossy-JPEG images.

The quoted numbers for JPEG are based on Huffman coding; with the alternate
arithmetic coding method, files are another five or ten percent smaller.
There's some doubt as to whether you can use the arithmetic coding method
without paying royalties to IBM, however.

-- 
			tom lane
Internet: tgl@cs.cmu.edu	BITNET: tgl%cs.cmu.edu@cmuccvma