Re: [Bug-ocrad] Library version of OCRAD

[Antonio Diaz Diaz]

Igor Filippov wrote:
> > As a first step, I have made ocrad recognize such characters.
>
> Hmmm, is the functionality different now? Could you give an example?

As I said in my previous post, ocrad was designed for full text pages. 
This means, among other things, that blobs touching the edges were 
removed as noise. For example, if you fed to ocrad this image:

P1
16 18
0000011111110000
0000111111111100
0001111000011110
0011110000111110
0111100000111110
0111100000111110
1111000000011100
1111000000000000
1111000000000000
1111000000000000
1111000000000000
1111000000000000
1111100000000100
0111100000000110
0111110000001100
0011111000011100
0001111111111000
0000011111100000

it would not produce any output.

I have now modified this so that for images with two or less lines, 
blobs touching the edges aren't removed. Ocrad can now recognize single 
characters like the 'c' in the image above.


> I still have to grab a portion of the image and extract the character shape out of it 
> - it doesn't seem to matter if I'm putting the result in Blob or
> Ocrad_Bitmap, what is the advantage of Ocrad_Bitmap?

Blob is an internal type of ocrad. It is neither documented not 
guaranteed to remain stable. Moreover, Blob has some requirements, like 
pixel connectivity, that could make osra inestable if your data does not 
meet them.

Ocrad_Bitmap is only an example of how an image can be safely passed to 
ocrad so that ocrad can scan it, create blobs, characters, etc, and 
produce a result.

> I also still plan to keep gocr in the loop, so I'm not sure how the
> current procedure could be made more efficient than it already is.

This is why I am proposing for input a struct similar to the one of 
gocr; to avoid making the current procedure less efficient than it 
already is.


> A dream feature request - what would be absolutely fantastic is the
> ability to detect characters which are touching a line (a chemical bond
> for example).
> I have a few sample images and none of the open source OCR software is
> able to deal with such characters. It is a very hard problem though.

Well, you can send me (to my private address) some of those images and 
I'll try to find a way for ocrad to recognize them.


Regards,
Antonio.