On 02/11/06, Steve Holdoway <[EMAIL PROTECTED]> wrote:
Hey,
If you're interested, bounce your ideas off us! If we can work out a new way of
fuzzy
matching on these images, then you'll be helping the OSS cause big time. I
haven't
manipulated images in this way in the best part of 20 years, so I'm a bit rusty!
Text recognition in images is hard. The spammers (the ones I've seen)
fuzz up the background a bit and have different coloured text to make
it even more difficult. There is a good reason why text-in-images
makes a good CAPTCHA (see wikipedia).
I would suggest some image transforms first to get the shape of the
letters easier to detect. Maybe convert to grayscale then increase
contrast. You're probabaly better off splitting up groups of letters
into words and then matching words rather than letters.
There's bound to be some research on this. Maybe ACM digital library
or something.
Cheers,
Carl.
