OCR to UTF8

On Mon, 21 Feb 2005 18:43:07 -0700, ray staggered into the Black Sun and said: snip

kooka doesn't do OCR; it calls a number of standalone programs to do that. Yeah, yeah, I know, but precision is important.

Hrm. Well, check this out:

Omnipage 10.0 commercial engine--

Business inventory is likely to expand moderately to take care of a rising volume of business. In no major lines are inventories out of line with sales. Residential construction has been among the weaker elements of...

Typereader 6.0 commercial engine--

Business inventory is likely to expand moderately to take care of a rising volume of business. In no major lines are inventories out of line with sales. Residential construction has been among the weaker elements of...

GNU ocrad 0.10 Free engine--

Business inveotory is likcly to expand modeately to take care o a rising voluloe o business. In no major lines ae inventories out oF lioe with sales. Residential con5truction has been amon6 the weaker elemeots oF...

...this is on a very clean image with no skew and well-formed characters. The commercial engines used were several years old; ocrad was the latest stable version available (0.10) according to portage. I'll try the other engines kooka uses--maybe tomorrow. Enabling spellcheck with aspell didn't help much--the thing wanted me to go through each mangled word and manually choose an appropriate new word for each one!

OCR is a difficult problem. I have some professional interest in it, and though I've never worked on an OCR engine, I have some familiarity with the problems and pitfalls involved.

Like another poster said, they can match addresses to a canonical database of valid addresses. That narrows the possible matches down considerably, making the problem much easier. No such constraint is present in the general OCR problem.

-- Matt GThere is no Darkness in Eternity-But only Light too dim for us to see Brainbench MVP for Linux Admin mail: TRAP + SPAN don't belong