Pattern recognition

[co2junkie_]

I am working with geometric pattern matching to detect circles, it works but it the results are unsatisfactory. Here are a few questions if anyone can kindly answer 

.

1) the circles have to be at the specific distance away from the webcam to get detected, can I get the circle detected for various distances at the same time?

2) Is there any filtering or preprocessing technique best for basic shapes or regular shapes ? 

My basic aim is to make a license plate recognition system, Im done with OCR with images, I have to now implement it in video acquisition mode, but I want the numberplate boundry to be detected by webcam too, will the geomatric pattern recognition work ? and one last question how to make OCR read text if the image taken is not from front (ie its angled) I tried to train the texts (angular ones) but didnt go well any idea there ?

I know these are too many questions, any suggestion or any slightest hint will even get me started 🙂

Thanks.

[Greg_S-Video]

Hello co2junkie_

Thank you very much for getting in touch with us!  Are you using Vision Builder AI or LabVIEW?

1)  You can get your circle pattern detected from varying distances away by setting the Geometric Matching setting to "search for matches which are scaled".

2)  You may be interested in some edge enhancement filters if you are finding the geometric matching is not working but is that the case?  You can decrease the score requirement to find a match and this may prevent the need to pre-filter the image.

You can certainly use a webcam to obtain your images.  Can you elaborate on your training technique for the angled letters?  I can attempt to reproduce this on my end.  For your interest, there is a tutorial on OCR of italicized characters.  This may provide some insight into your situation.

Optical Character Recognition (OCR) with Italic Characters
http://zone.ni.com/devzone/cda/tut/p/id/5881

Please let me know of your results and how I can be of further assistance.

[co2junkie_]

Thanks for replyng.. 

Yes Im using Vision Builder AI 

Geometric pattern is working, but the recognition is not efficient.. and here is why i think its not working fine:

The camera i am using for taking picture (for template editor) has a resolution 3648 x 2736 while the webcam I use for the video acquistion has pretty low res (1366 by 768), when training the image in template editor i get pretty good results using the edge based algorithm and slight modifications with threshold settings, but when i use i webcam for recognition the results aren't good (webcam looses the detail), and if i try training the image in template editor shot by webcam, I don't get good edge result, since the image is blurry, or more like 'homogenous' ,

can this be the reason of not having efficient recognition ?

Yes, by angled image, i meant the image is taken at certain angle and not from front. Like I take image of a license plate, from sideways or from above.

and thanks for the varying distance idea it workd well 🙂

[Greg_S-Video]

Hello co2junkie_

Thank you very much for yout response.  Let me know if I am interpreting the situation correctly.  The results seem undesireable when training with one camera and capturing with another.   What happens when you train and capture with the same camera?  Are you able to decrease the match score in order to achieve better results?  Please let me know!