Re: [Bug-ocrad] A few ocrad problems

[Don Moir]

Hi Antonio,

> > 4) Failure to detect merged ti, vi, im, ll,  in merged_ti_vi_im_ll.pbm
>
> This one is the most difficult. The last version of ocrad has fixed some problems like those, but there are lots of them (even 
> with more than two letters merged). I'll try to fix as many as I can, but I don't promise anything.

I am pretty new to OCR, but I tested several OCR programs before I tried using ocrad. Now for my purposes ocrad has worked better 
but the space problem and merged characters are a problem (not just for ocrad of course).

I like the idea of feature extraction, but I am thinking I would like to generate the feature/rule set either on the fly or in 
database files. It seems to me that this could aid in the detection of merged characters. Like we absolutely know what a cap T looks 
like and a merged TT is just 2 cap T's together. I don't know about this as I have not looked into it to much. But some character 
sets may be like old english style and those might be hard to recognize with a fixed rule set.

Hopefully I will eventually have the time to look at it in more detail.

Thanks,

Don

----- Original Message ----- 
From: "Antonio Diaz Diaz" <address@hidden>
To: "Don Moir" <address@hidden>
Cc: <address@hidden>
Sent: Sunday, June 02, 2013 3:46 PM
Subject: Re: [Bug-ocrad] A few ocrad problems


> Hello Don.
>
> Don Moir wrote:
> > Results so far look good and better than other other sources I have tried.
>
> Thanks for the feedback.
>
>
> > 1) An orphan capitial letter I fails to be detected.
>
> Textline::recognize2 is supposed to become some kind of expert system for post-processing of recognized text, but it yet lacks a 
> lot of "rules". This is one of them. I'll add rules for isolated 'I' and 'UP' in the next release of ocrad.
>
>
> > 3) Failure to detect a space character in latin_space.pbm.
>
> This one is a little trickier. Currently ocrad measures the distance between "character boxes". It should measure the distance 
> between the black blobs inside those character boxes, but this is more difficult to do. I plan to fix this in a future version of 
> ocrad.
>
>
> > 4) Failure to detect merged ti, vi, im, ll,  in merged_ti_vi_im_ll.pbm
>
> This one is the most difficult. The last version of ocrad has fixed some problems like those, but there are lots of them (even 
> with more than two letters merged). I'll try to fix as many as I can, but I don't promise anything.
>
>
> > The attached zip contains 6 files:
>
> Next time, please, send the images to my email address, not to the list. Thanks.
>
>
> > I am wondering if possible merged characters should be added as special characters. like TT, ti, etc so then in future it's easy 
> > to add such combinations.
>
> I have in fact removed some such combinations from the last version of ocrad. There are just too many of them and trying to 
> recognize them worsens recognition results.
>
>
> Best regards,
> Antonio.