Poor OCR results

[SteepleChase]

Some document's OCR results are excellent. This one's are not. Why? What would help?

I suspect that the original document is too poor quality.

[Willy Van Nuffel]

Hi SteepleChase.
The OCR functionality is a nice piece of software that has been add to the PDF-XChange Viewer Pro.
I did several tests with it and did find out that the "Accuracy" parameter does not really do what it should be supposed to do. When you set it on "Low" or "Medium" the result is fairly good. When you set it on "High" the result is rather bad.
A test with your "test document.pdf" confirms this.
So, to the people of Tracker-Software, if you could make something out of it that results in the best combination of "Low" and "Medium" accuracy, it would almost be perfect. Already thanks to all you for the effort that has been done !

[Paul - Tracker Supp]

Hi we have the document and I'l pass that on to the OCR lead developer when he returns from his Christmas Holidays next week. We'll provide feedback here.

hth

[wilfriedh]

Just for interest I ran the "test document.pdf" trough PDF-XChange Viewer's OCR (each of the three levels) and Abby FineReader 9. You can see that also FineReader's results are not perfect. The original is too blurred.

Wilfried

[Paul - Tracker Supp]

HI wilfriedh,

thanks for the input. That's quite interestinbg to see. Finereader did do a better job but it is also a $ 169.99 product that has had years of market place trial. This is a free OCR that is in it's first release.

We will still be working on this. hth

[Walter-Tracker Supp]

Hi,

The recommended accuracy setting for most document is "medium"; in some cases the trade-off between speed and accuracy makes it worthwhile to use "low", which is faster but slightly more error-prone. High accuracy should generally be used for high resolution documents with small text; for general use with typical scanned documents (letters, forms, etc) it may end up performing worse than medium. We have left it up to the end users to determine which method is best for their specific document.

Also, the input document you provided is fairly low resolution. If you "zoom in" you can see that there are a lot of things that typically cause problems for OCR - poor delineation of letters and letters often contact their neighbours quite significantly. A higher scanning resolution should resolve this.


We are continually developing our OCR functionality and our products in general, so your feedback is greatly appreciated.

-Walter