I was quite sure that FuzzyOcr project is dead, because a few
months ago I was trying to contact his author, Decoder,
but no success. Probably he was very busy :) Fortunately, it seems
He is very busy getting an advanced degree. He still manages to put out the
occasional patch, and several others have done quite a lot of work on it.
I've found a threat about rotated spam images at FuzzyOcr page [1].
Currently Decoder hasn't time to implement checking image rotation,
but he will try to do it in the future. Now we can only work-around it,
for example using the preprocessor/scanset settings.
Who of you do rotate images in your FuzzyOcr? Do you use fixed
degrees or detect the skew angle and rotate the image accordingly?
Could you share this?
I don't personally check for rotated images, since all my image spams get
quite enough points from other things, so I don't need to make the extra
effort.
There was someone else a couple months ago that has a fairly long thread in
the mailing list about his experiments with rotated images. Unfortunately I
didn't save any of those in my local archive. But as best I recall, he was
suggesting multiple scan sets at (I think) about 8 and 18 degrees each way.
I remember there was some talk about rotations over a certain angle being
difficult to detect, but I don't recall the exact details now. I think it
was problems with font distortion doing the rotation, and someone else had
some suggestions to get around that problem.
All this is a little hazy in my memory. This was not too long after images
spams started and they started trying to avoid OCR detection. They started
doing rotated images, but it didn't seem to last too long. I'm guessing
that they probably didn't get good hit rates from the spammer's "customers".
Loren
