RoboRealm Forum

No Title

Tue, 30 Nov 1999 00:11:00 EST

OCR module just underwent an upgrade to become rotationally invariant, i.e. it can read up-side-down text. If you download the latest version you should be able to run the attached robofile to get the results shown in the attached image.

Note, there are still some errors. Namely the 5 and S get confused since the characters are very similar. These errors can be dealt with depending on the nature of your application. If you know all the part numbers, then adding them into a word database and using that will further improve accuracy.

You will need at least 10 more of these types of images to really test if the system is stable and reliable.

STeven.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Hi Steven,

Any updates on the previous question?

No Title

Tue, 30 Nov 1999 00:11:00 EST

Sure..Thanks again.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Roshan,

Thanks, I think we understand the setup now. Give us a couple days to try a couple solutions out to see which one might work best.

STeven.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Yes, the tablets are placed in holding cavities( holes drilled on a disk at equal distance from each other) to hold the tablets in one place so that they don't move or topple, so the position of the tablet would be the same all the time.

Its just that round tablets can rotate freely inside their holding cavity during movements. Now, if the tablets were in oval or rectangular shape this problem wont arise.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Are tablets always in the same position? I.e. they all seem equidistant from each other. Does the machine always place them close to where they are currently now?

I'm trying to understand what expected movement you would expect to see relative to each other. For example, if you took another picture from another set of tablets do the tablets overlap for the most part?

No Title

Tue, 30 Nov 1999 00:11:00 EST

Hi Steven,

Please refer to the attached image. This is the actual image from my machine setup. You can see that all tablets are at random angles.

And thank you for making time to help me out here.

No Title

Tue, 30 Nov 1999 00:11:00 EST

I was just using the second image you sent that showed only one tablet. The assumption is that one tablet is in view at a time ... you can use the Crop function to ensure that this is the case.

If your setup does have more than one tablet and they are always relatively in the same position you can use the Split_Image module to break the image into two successive images that get processed.

Again, one image will not be enough for testing purposes.

STeven.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Dear Steven,

It doesn't seem to work. It only shows the result of the first tablet(Please refer image). Also upgraded to the latest release.

Can I try to create an image database of these 5 characters in different angles and use them as a reference database, will it work? OR is it possible to use image verification technique for each character?

No Title

Tue, 30 Nov 1999 00:11:00 EST

The attached seems to do the trick... you can change the rotation using the first module and ensure that the text is still read correctly regardless of orientation.

Note, one image is never enough for testing purposes!

STeven.

No Title

Tue, 30 Nov 1999 00:11:00 EST

HI Steven,

It can be rotated to any degree, I just used the extreme in the example image. I'll explain the scenario : This issue of rotated tablets happens only with round tablets, since after printing when it moves ahead, due to the jerky movement of the tray and machine vibration, the tablet rotates in its holding cavity. Its a free rotation, can be any degree.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Roshan,

Quick question, do you expect the text to be rotated in just those two ways (right side, upside down) or is it possible for the text to appear at say 33 degrees?

The reason for my question is that determining upsidedown text is easier than any rotation.

STeven.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Hi Steven,

Thank you, it works great. Just one more issue. Is it possible for the module to detect words which are rotated? I tried but it reads it incorrectly.

Please refer to the image.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Roshan,

Thanks for the better image. That appears to work well. I've attached two robofile that recognizes both images that you posted. It turns out there was a problem with the cached image file (OCR\\database.dat) that was released. You will have to download the latest version of RoboRealm to have either of these configuration files work as expected.

v2.62.2 has this fix.

Let me know if you have any further issues.

STeven.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Hi Steven,

Actually im trying to read alphanumeric text from a printed tablet. My goal is to identify misprinted tablets and send a signal to the PLC for rejecting such tablets. Since the tablet diameter is around 8mm, the print on it would be about 5mm. To get a descent resolution i'm using a canon 1100D DSLR as a webcam(sparkocam app)

The font that I'm using is Arial Unicode MS.

Attaching a high resolution image of the whole tablet.

No Title

Tue, 30 Nov 1999 00:11:00 EST

Do you have a higher resolution version of that graphic? Or is that the largest image you have?

What you want to do is to use something like Adaptive Threshold with the letters set to white before you use the OCR module. But the above graphic is a bit small and thus difficult to match.

Also knowing the font type might help to create a better match.

STeven.

OCR

Tue, 30 Nov 1999 00:11:00 EST

Hi,

Can someone please send me an example of using the OCR module in roborealm...I tried playing around with the OCR module, tried converting the image to grayscale, inverting etc but i'am still unable to figure out how it works. Have attached a sample of the image i'm working on.