[Bug-ocrad] Re: improving discrimination between i and j

[Antonio Diaz Diaz]

Hello Lode. Thanks for your interest in ocrad.


> I wonder if you are taking patches to ocrad...

Any useful collaboration is well received. Thanks.


> This rule improves the detection of "j"
> Maybe it would be better to have features "has_descenders" and 
> "has_ascenders"
> and test add_guess(f2.has_descenders()?'j':'i',0)
> where has_descenders is related to the number of pixels below the 
> baseline of the character.

The "has_descenders" feature is not a feature of the character itself. 
It depends on its position in the line. Ocrad uses the "baseline" 
(charbox.bottom()) in a number of places, but the algorithm that 
calculates it has to be improved more.

Could you send me (to me, no to the list) an example image for every 
patch you just sent? I verify every change with a rather large test 
database before deciding if it is worth to inlude it.


> I'm trying to understand the following piece of code (character_r11.c:124-146.
> Could someone describe this in words?

It is a lame atempt to recognize a merged 'fl' that has been recognized 
as a 'n', by trying to detect the 'f' horizontal bar and the 'l' lower 
serif.


> Actually, it would be nice to have a one-line description of all the functions
> like profile.minima profile.increasing etc..

Yes, it would be nice. I try to comment the code but, 1) I am lazy, and 
2) sometimes I don't know how to do something until it is done. ;-)


Regards,
Antonio Diaz.