ocr character file causes memory leak during OCR. Please can someone help?

[markward]

I'm trying to use OCR to read characters from an LCD screen.

It all works 🙂  but... it leaks a LOT of memory, becoming unusable in a loop.

I think the problem is the character file. I've tried a few ways to create one (e.g. from within vision assistant, from within the standalone training utility, and with- and without automatic thresholding), and none of them fix it.

The NI OCR example vi works fine using its own abc file. But if I use any of mine there's a memory leak each time it runs, typically of a few hundred k bytes.

Please can anyone help or say what's going on? Or fix the abc file?

Thanks VERY much.

[markward · Accepted solution]

Tech support in the UK have come up with a fix. If they give permission, I'll post it here.

The problem may affect any OCR training file made using the GUI tools in release 2013.

[Adam.B]

I worked with Mark on this issue, as he mentioned there is a possibility that ".abc" files generated using the 2013 version of either Vision Assistant or the GUI OCR trainer could cause a memory leak when repeatedly used with the OCR Read VI. ".abc" files generated using the standard OCR Train VI don't appear to be affected. NI is aware of this issue and a report has been filed.

I have created a VI that uses the standard VIs to generate a ".abc" file, The VI allows you to train a character at a time by drawing a box around it and entering the character value. Mark has used this VI to successfully generate ".abc" files that do not create memory leaks so hopefully this will help anyone in the same position.

[markward]

We've just used it and can confirm that it works. The only drawback is that to avoid the meory leak I had to use the low-level OCR vi when reading too, not the gui version.

To summarise, to avoid the memory leak we had to

* make the abc file using low-level vis

* do the ocr using low level vis

using the gui versions for either step causes the problem. Hopefully this will be fixed in labview in future.

I couldn't get the low-level ocr vi to read multiple lines in one go, but otherwise it was fine.

The application is reading numbers from an lcd screen in real time.

[Silver_Shaper]

Dear Adam,

Any updates on the above issue?? . Me too facing the problem. Its difficult to train the characters using Low level VI's. 

[Adam.B]

Hi Sasi,

Unfortunately there are no further updates at this time but the bug report is still open so R&D are investgating.

[Silver_Shaper]

Hi Adam

Thanks for the reply. I hope it will be solved iin 2014 Version.

[muks]

Hi Adam,

Can you post the CAR number?

[Adam.B]

Sorry, I should have posted this earlier, the CAR number is 456495.

[muks]

Thanks