Newsgroups: comp.graphics,sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!howland.reston.ans.net!vixen.cso.uiuc.edu!uwm.edu!msunews!harbinger.cc.monash.edu.au!news.cs.su.oz.au!metro!extro!maxtal
From: maxtal@extro.ucc.su.OZ.AU (Tim TAL Lister)
Subject: Re: Unhalftoning Halftones
Message-ID: <D8CEDy.HJ@ucc.su.OZ.AU>
Sender: news@ucc.su.OZ.AU
Nntp-Posting-Host: extro.ucc.su.oz.au
Organization: MAXTAL P/L C/- University Computing Centre, Sydney
References: <3o756l$eot@mack.rt66.com> <1995May3.130026.1020@csginc.com>
Date: Wed, 10 May 1995 03:15:33 GMT
Lines: 57
Xref: glinda.oz.cs.cmu.edu comp.graphics:76206 sci.image.processing:14507

In article <1995May3.130026.1020@csginc.com> bob@csginc.com writes:
>In article <3o756l$eot@mack.rt66.com>, cj@rt66.com (Richard Carter Jr) writes:
>> I'm looking for a program, preferably Windows or DOS, that includes in 
>> it's functionality the ability to transform the dot patterns of halftone 
>> photos (b/w and color) into a continuous-tone digital image (or close) so 
>> that standard image filters (sharpening, edge detect, etc.) can be applied 
>> to try and enhance the image even more.  
>> 
>> I've done what I can by hand using only the image filters available in 
>> various commercial- share- and free-ware packages.  The results have 
>> usually been disappointing.
>
>I've some doubt if it is even possible to reconstruct the original.  What
>process have you tried?  I've double the sample size, blurred, then halved 
>the samples on paper maps with ok, not great results.  The approach seems
>rather naive.
>
>> 
>> I'll look into any suggestions -- though the less expensive, the better 
>> (isn't it always the way?).
>> 
>> Thanks.
>> 
>> -- 
>>  CJ
>> cj@rt66.com
>> * It's hard to reach for a star and still keep your shirt tucked in. *
>> 

  an interesting problem.  the core of this problem is that the half-toned
imaged is 1 bit deep, and most of the original grey-scale data has already
been thrown away by the half-toning process.

  You might get somewhere if you could identify the exact h-t process used.

  E.G; if you knew that the process was error diffusion, mono-directional
Stucki dithered, you might be able to create an inverse mapping that
collected proportions of neighbour pixels according to the Stucki weights,
and generated a grey-scale value for each pixel.  This would only be a
rough approximation, however, and astonishingly bad artefacts might be
expected in places.  If you didn't mind hand editting the artefacts, this
might be OK.  Of course, if you can't identify the method, this is hopeless.

  you might be able to write a package that tries all sorts of methods, and
pick the best, but ultimately this problem is not solvable in a generic sense,
since information discarded by the half-toning process is, well, discarded,
and thus not accessable.

  sorry to rain on your parade,
				regards,
					tal

-- 
;----------------------------------------------------------------------
        TIM (TAL) LISTER,         maxtal@extro.ucc.su.oz.au
	Maxtal Pty Ltd, 81a Glebe Pt.Rd, Glebe, NSW 2037, AUSTRALIA
;------ Graphics & C++ libraries, Games & Image Processing ------------
