To: OSHA
52 posted on
09/26/2004 8:26:35 AM PDT by
Scutter
To: Scutter
If I could get a digitized (any resolution) picture of a sample of characters in the "documents" and equivalent pictures of putative typewriter font characters and MS Word characters, then I have a method (currently under review for publication by IEEE) for measuring similarities between shapes.
For, example, given an "A" from the document (several would be better) and an "A" from some font sets, I could tell which were closest. Even better, I could do it with combinations such as WA (or anything) to demonstrate kerning differences.
I actually get a metric between shapes. Zero distance implies identity and there is a triangle inequality. The method is rotation and translation invariant (means I don't have to register the images.) It is not scale invariant (however images can be presaged) but pictures at different resolutions can be compared. I should get good comparisons modulo the noise level in the pictures.
60 posted on
09/26/2004 8:39:08 AM PDT by
Doctor Stochastic
(Vegetabilisch = chaotisch is der Charakter der Modernen. - Friedrich Schlegel)
FreeRepublic.com is powered by software copyright 2000-2008 John Robinson