Re: [Bug-ocrad] Technical documentation summary readme.txt, page skew, Hough transform.

[Antonio Diaz Diaz]

Chris K. Skinner wrote:
> As you are probably aware, there are software patents in some countries.

Yes, in almost every country with a corrupt and/or fascist government.


> If you had some kind of outline of the algorithms that were applied per 
> version of the software that would greatly help someone new coming in 
> fresh off the street to gain a quicker understanding of stuff in 
> general, and probably demonstrate to the world at large that you have 
> invented something new that could not be patented / stolen / claimed by 
> some greedy corporate dudes.

I sympathize with your idea, but I lack the time and the ability to 
explain the algorithms I use or invent. On the other hand, a patent is 
valid even if I invented it independently, so it won't be an effective 
defense.


> Do you have any design notes, bibliographic citations, web links to 
> information that you've made use of , release notes for what algorithms 
> are being used / abandoned.

The short answer is no. I have looked into the source of gocr and 
claraocr, but I haven't got anything from them. I use the Otsu algorithm 
for binarization (as gocr does). Apart from this, I work mostly in a vacuum.


> In the J. R. Parker book w/CD ROM "Algorithms For Image Processing And 
> Computer Vision" that I have read, the author provides a couple of 
> algorithm suggestions for combating the page skew angle issue.  A 
> Hough-transform when applied to the dots of the bottoms of the bounding 
> boxes of glyphs results in a page skew angle in degrees (with his source 
> code, that is).

This has a number of problems:
- Hough-transform, and in general any transformation on the whole image, 
is slow as hell.
- What if the line is not skewed but curved? (frequent in scanned books).
- "The bottoms of the bounding boxes of glyphs" are usually not aligned.
- etc...

This is why I expect working code, not suggestions, from possible 
collaborators. (Show me the code, you know?) ;-)

By the way, ocrad's algorithms are designed to be resistant to page skew.


> Another approach is to use angle-independent Complex-Number-Coefficient 
> Neural Networks to use as feature recognizers.  The Japanese promoter of 
> these neural networks says that they are Affine-Transform insensitive, 
> and thereby can recognize a pattern that has been so transformed.

I would like to see this recognizing a page in less than five minutes 
with good accuracy.


> This too is just a theory.  I don't have a copy of any books on 
> Complex-Number-Coefficient Neural Networks, or any source code from a 
> competent mathematician who has converted the advanced mathematics into 
> working C++ code examples.

Don't worry. "Advanced mathematics" are the sofware of the future... and 
they will always be. ;-)


Regards,
Antonio.