How to create Character Set Files (without using Training)?

[vik1]

Hi,

     I am using NI Vision 8.5. we are evaluating its OCR capability and trying to replace it with an open source version TOCR. My question is say I have an image that has a standard fonts like Times New Roman. How do I get OCR to read those characters without training it? I am thinking that I have to use a character set file. Where can I obtain a character set file?

    also, another question which I have is about the precision required for ROI. right now, TOCR requires those characters to be exactly in the same position and 1 or 2 pixels off throws it in a dizzy. Does Vision's OCR have the same problem? If the text is within a decent ROI will it be able to fidn it?

Thanks

vik

[BruceAmmons]

The NI OCR is designed so that you train whatever fonts you need.  It isn't too difficult unless you have a lot of fonts to train.  In this case, you could generate lines of text in an image and automate the training process.

The ROI for the NI OCR is fairly flexible, as long as you don't pick up any extra marks or a surrounding border.  Extra blank space doesn't affect it at all.

Bruce

[risheeb]

Hi Vik,

Thank you for posting on the National Instruments discussion forums.

In order to configure NI Vision to properly identify characters, you must train it; there is no way to use a character set file to do this automatically.  Training NI Vision to recognize the Times New Roman characters should be fairly straight forward though.  As Bruce mentioned, you will simply need to generate a line of text to train the software with.

Regarding your second question, NI Vision's ROI feature is much more flexible.  It should recognize characters in various locations, although you may need to rotate images if the characters are oriented differently.

I hope this helps!


Rishee B.
Applications Engineer
National Instruments

[vik1]

Rishee,

            Hi! Thanks for replying. Is there any way you can send me some example vi's that do OCR training? I am just getting into this and my eval purchase expires in 2 days. I need to convince upper management to buy NI Vision and  they are happy with the Pattern matching but OCR would make it a cinch. I have attached a sample example file that I am using to let it train. Even if you can point me to somewhere I will be happy.

Thanks

bhavik

[risheeb]

Hi Bhavik,

NI Vision includes the OCR Training Interface, which is used to develop a Character Set file from your images.  The reason you should use this to create the Character Set instead of a Character Set file that has already been made for a specific font is because you want to create the Character Set file from images which are as similar as possible to the conditions under which you will be searching for characters.

A link to open the OCR Training Interface should be available in the National Instruments -> Vision folder on your Start Menu.

Once you create the Character Set, you can use the OCR VI's in LabVIEW to begin recognizing characters.  There is a shipping example called "OCR Example" which demonstrates this.  To access, open the Example Finder in LabVIEW, then navigate to Toolkits and Modules -> Vision -> Applications.

Please reply if you have any difficulties in using the OCR Training Interface or the LabVIEW example.  Thank you.

Rishee B.
Applications Engineer
National Instruments