best font face / size for optimal recognition

Hi,

We generate some documents ourselves on which we provide a code which enables us to classify the document afterwards (we don't use barcodes as these seem to take more place and less information on a restricted area of the page). Could you suggest which font face / size we use best then to have optimal recognition when scanning @ 200 dpi?

thanks
matthias

Hello,

To get good OCR results, the text must be clear and readable. Also, the size of the image should be A3 or smaller.

Thanks,
Maen Badwan
LEADTOOLS Technical Support

You mean there is absolutely no difference between Arial 20pt and Times New Roman 8pt? kinda hard to believe ;) There must exist an optimal, or at least, a minimum font size with maximum performance. Also, I guess there must be a difference in recognition between serif and sans serif fonts?

Hello,

The toolkit and OCR engine documentation do not contain published information about a minimum font size, or specific font type for optimal recognition. Of course the larger the font and the higher the resolution, the better OCR results you will get.

I suggest you do your own testing using the restricted area in your documents to see what type of codes to use.
I also recommend using a lossless compression format, such as CCITT G4 or JBIG if your images are black and white.

Thanks,
Maen Badwan
LEADTOOLS Technical Support

we are scanning 200dpi A4 sheets in color and saving to JPEG (no negotiation there) and since we are still in the design stage of the forms we can choose the place and size of the form identifying string we want to recognize. Of course, the smaller this area, the better because then there is more space left for form input. I feel a bit hesitant to rely on the results of a rather limited test set I am able to generate and test myself since I cannot be sure if I was lucky or not in my specific case to have a correct recognition. But thanks anyway :)

Hello,

Using JPEG is a problem in itself.
There are 2 groups of JPEG formats, one of them is lossy and the other is lossless.

The problem with lossy JPEG is that it changes the pixel data every time you load and save. This could lead to more recognition errors when using OCR.

The problem with lossless JPEG is that very few non-LEAD applications understand it, so unless your images will only be viewed by your own application (or another application built with LEADTOOLS), you should generally avoid it.

Is it possible to save the data you need in the JPEG comments instead of writing it on the image pixel data?

Thanks,
Maen Badwan
LEADTOOLS Technical Support

Hello,

did u get any standard font size and face as I am working on OCR for one of my utility and williing to know if LEADTOOL OCR is performing well using specific font.

Hi Bhavik,

we ended up using the OCR-A font [1] which claims to be designed specifically for OCR processing.
However, it turned out this only works well if the zoning engine creates a separate zone for this text field which does not always happen, i.e. when the text is aligned with other text (in another font).

So, the best recommendation I can give is: use the same font as any other text on the form. There are certain limitations on the font size, but I don't remember them off the top of my head.
As long as you stay within 'normal' size ranges (> 10) and standard font faces, you should be safe.

HTH
matthias

[1] http://en.wikipedia.org/wiki/OCR-A_font