Re: river detection

[Hans Hagen <pragma@wxs.nl>]

At 01:45 PM 1/9/01 -0500, Ed L Cashin wrote:
>Han The Thanh discusses in his thesis several problems that computer
>typography systems face.  One problem that no current systems address
>is "rivers", visual lines of blank space like this:
>
>
>     XXXXXXXXX  XXXXXXXXXXXXXX XXXXXXXXX XXXXXXXX
>     XXXXXXX   xxXX XXXXXXXX Xxxxx   XXXXXX XXXXX
>     XX  XXX   XXXXXXX XXXXXX XXXXX XXXXXXXX XXXX
>     XXXXXXX  XXXX XXXXXXX XXXX  XXXXXXXX XXXXXXX
>     XXX XX   XXXXXXX XXXXXX XXXXX XXXXXXXX XXXXX
>     XXXXX   xxXX XXXXXXXX Xxxxx   XXXXXX XXXXX X
>     XXXX   XXX XXXXXXX XXXX  XXXXXXXX XXXXXXX XX
>
>... where there's a river on the left.  Hans has been doing many
>experiments combining MetaPost and TeX, where they share more
>information about the visual appearance of the typeset text than has
>traditionally been available.
>
>Would it be possible to use MP and TeX together in order to detect
>rivers?  If so, it would be a computer typesetting first, according to
>Thanh's thesis.  :)

I must say that my first thought was that you might had a point, but after
half an hour metaposting [playing a bit with picture postprocessing, of
which you can find an example in the metafun manual] i think that, given
that there was a good method, it could as well be done in tex itself, since
tex has as much knowlegde in this respect as metapost, i.e. the boundingbox. 

What i did was (1) converting text into pictures, (2) converting
boundingboxes of chars into matrix points and (3) looking at the result.
Maybe some matrix guru could program an ananalyzer but my math is to weak
for that. 

I think [but thanh may disagree] that grayness is something perceptual and
rivers are things recognized by our eyes and brain at a quite low level,
not so much analytical. So, if there was a way that tex could send an
paragraph shape in terms of boundingboxes to a file, and after that a
separate process could feed that into a neural net [optionally converted to
bitmaps so that the character shape could be taken into account], and the
net could send back a badness value to tex, so that there could be an
additional pass ... 

I think that it's not that hard to extend tex with a spawned process in the
paragraph builder and let it act upon the baddness. The main question is:
how do we convince thanh to provide that hook, and after that, how do we
trick ed in writing that analyzer. 

But this is a nice thread, 

Hans   
-------------------------------------------------------------------------
                                  Hans Hagen | PRAGMA ADE | pragma@wxs.nl
                      Ridderstraat 27 | 8061 GH Hasselt | The Netherlands
 tel: +31 (0)38 477 53 69 | fax: +31 (0)38 477 53 74 | www.pragma-ade.com
-------------------------------------------------------------------------