OCR technique not working to detect the logo

[பரத்_குமார்(bharathkumar)]

Hi guys,

I am trying to detect the text of the Logo on the product in an assembly line. I am failing to detect it in all the products.

First time I placed it in the ambient lighting conditon acquired the image and processed it. It worked in Lab @ 30 - 40 lux of illumination and failed in the assembly line where the illumination varies from 50 - 130 lux whole day(24hrs x 7 days). Based on obseravation I tried to vary the threshold level by manual for various lux values that too failed.

I decided to limit the illumination partially on the logo to maintain at a particular level it is shiny on the ends and am not able to read the end letters  but the center letters are readable.

One thing am not able to do is convince my seniors to buy diffuse light setup for  the inspection. Is there any alternative method to solve this?

Thanks for your help in advance, I have attached the image with this post

Regards,

Bharath

[BruceAmmons]

This is not an OCR problem.  You need OCR when the text changes, and you need to read the text.  Applications might be reading serial numbers or odometers, etc.

Since this is a fixed logo, you could use pattern matching, with the entire logo being a pattern.  That would handle most light variations, I suspect.

What do you actually need to do with this?  Verify the logo is present, or something more complicated?  That makes all the difference in the world.  For example, if you need to verify the details of the logo, you would probably use a golden template comparison after the pattern matching.

Bruce

[MangeshD]

Is there any specific application that you are trying this for. I mean if it is simply a one logo kind of thing you can go with Bruce's answer but if you have to check characters and it is frequently changing then train character set and do teh OCR.

You can refer examples shipped with NI LabVIEW for
1. Pattern Matching
2. Golden Template Match
3. OCR

[பரத்_குமார்(bharathkumar)]

Hi Bruce,

Thanks for your suggestion. Initially it was done with pattern matching. But the requirement changed to read the characters i.e check whether all the characters are present and look for tilted characters. About the golden template matching I shall give it a try. Update the status later.

Thanks,

Bharath

[பரத்_குமார்(bharathkumar)]

Hi LvKraken,

Thanks will try the golden Template matching.

Regards,

Bharath