[FoRK] image processing for OCR

Steve Nordquist < signa at everestkc.net > on > Tue Jul 4 03:37:15 PDT 2006

>> The algorithm is called a flood fill.

{And, we cry drily; order up a rescan!}

...and a LINK is provided?!  It's...  Well, Xara's is supposedly special,  
but that's not REXX there in the 'native format' XML, is it?
Early ancestor to the earthquake fill, the mosaic fill,

> I'm familiar with it, and unfortunately it wouldn't do.

Flood fill with variable threshold for punching through?  Fantastic thing.
Using a threshold and then doing some handy logical operation on the
original and the result ought to help, surely?

The pseudo-2D and 2D masks are the general set of what you've asked for;
blobify (min, max), floovex, bignoise, stroke extraction (stacatto);
extract 20-24pt. (lousy scan of a) yellow outline font and crank that into
stark relief, remove all the 2-pixel small-font poesic conincidences,
OR the images or something and go to the OCR pipeline.
	Failing that, glyph raster matching!

Yes, Kai's AfterEffects probably has something in it, but this runs
into the already stated problem of asking Adobe if it wants to own all
your base....

And speaking of the secret recipie for Texmexicum-211;

	I could suffer some more of those restaurant reviews, too!
That was so much better than the service-dogs-at-the-lobster-pot story,  
and the plot is afoot to make it ersatz without really even paying  
attention after a thorough reading and wild blind guesses at wine  
pairings.   8 course limit unless the local Asian Festival does a 30-hour  
food marathon thing.  (They even told me how to do the vegan sausage  
thing, so I've few excuses then.)  What's with the pinot crop of tomatoes  
with seeds already sprouting in the fruit, anyhow?  Leave that farmers'  
market alone from now on?  Call them varietals?




More information about the FoRK mailing list