Help reading LED display

[Fad]

Hello,

I am using IMAQ to read the number displayed on a LED Display. I built a program based on the given example of read LCD and another one based on the OCR example but unfortunately none was a success 😞

I must be missing something in here and any help would be really appreciated.

I am attaching the Files and a sample image of the display.

Thanks

[wbadry]

Hello fad,

can you try to enhance a little bit the focus as the image is blurry. Moreover, your camera is not prepenicular to your LEDS and that's why LCD reading fails. You need to perform image correction for the projection your camera is positioned in. As letter "A" in first Character would have different characteristics than letter "A" in the last character.

Hope it helps

Have a lovely day

[AnalogKid2DigitalMan]

Could OCR be failing due to the discrete dots of a character being separated?

Actually I was thinking that a little blur may be of use here, either by putting camera out of focus or using a thin piece of frosted glass over the display.

OCR may not know what to make of the zero with the slash in it.

Can OCR be 'trained' with a unique character set? I have not worked with OCR before.

-AK2DM

[Vladimir_Drzik]

To use OCR, the character dots have to be connected, so that each character is regarded as one "blob". I'm attaching a simple preprocessing Vision Assistant script to demostrate how to do the connecting. It also might be a good idea to do some geometrical image correction, as VBCoder says. However, I think the OCR algorithm is quite robust against such a degree of image deformation, so the correction might not be necessary.

As for the training, the OCR does not know any prelearned characters, so it's not a problem to learn any characters the LCD panel produces. However, if you do some preprocessing before feeding the image into OCR (you'll probably have to), you also have to do the learning on the preprocessed images.

Message Edited by Vladimir Drzik on 09-07-2009 12:58 PM

[BruceAmmons]

Actually, the dots don't need to be connected.  There is a spacing parameter in OCR that specifies max space between dots in the same character.  The main issue is that the spaces between the characters has to be larger than the spacing between the dots of the individual characters.

Training a template in multiple positions would also overcome the sizing issues.  It would be better to look at the display straight on, but if that isn't possible you could train the same character in the leftmost and rightmost positions.  If it is a really long display, some middle positions could also be used.

Bruce

[Fad · Accepted solution]

Thanks everyone for the quick support.

The best way that i figured out to solve this problem s by using OCR and training the characters. evenmore, i had to make an ROI for each number and recognize it alone.

my problem is basically solved until now 🙂

Thanks again

[wbadry]

Good Job Fadi