imgseek appropriate for automating inspection?

[Tim Harrison]

Hello! I'm researching software that could be used to help me automate inspections in a manufacturing process.  I'm hoping the imgseek users could tell me if imgseek could be appropriate.

We need to inspect thousands of small electronic LCDs for quality problems.  E.g., backlight does not come on.  Or a segment (part of a letter / number / icon) is missing.  I will have a "golden standard" photo of a good product (lcd).  Then I need to compare thousands of other photos to the "golden standard" and try to detect any problems.  So I'm looking for very slight variations in similarity.  Could imgseek be appropriate?

Thanks for any insights you might provide.

-Tim

[Ricardo Niederberger Cabral]

Hi Tim. I guess so. Some testing is required though. Have you tried integrating isk-daemon (http://server.imgseek.net/) to your production process? If you need help doing so, please let me know.

Or send me some image samples and I'll run some tests.

Regards,

Ricardo

--
You received this message because you are subscribed to the Google Groups "imgseek_users" group.
To view this discussion on the web visit https://groups.google.com/d/msg/imgseek_users/-/n6nVv9VZobEJ.
To post to this group, send email to imgsee...@googlegroups.com.
To unsubscribe from this group, send email to imgseek_user...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/imgseek_users?hl=en.

[Tim Harrison]

Hey Ricardo.  Yeah, my first step is to run a bunch of proof of concept tests.  What kind of approach would you recommend in terms of building a database and running tests?  From what I can tell, I should start by adding one "good" product photo.  Then I would add the test images one by one and call calcImgDiff.  Something like this:

1) Add test image to database

2) testDiff = calcImgDiff(dbId, goodImageId, testImageId)

3) if testDiff > someThreshold then flag as "bad" product

4) Delete testImageId from database

5) Repeat

Does this seem like the right way to do this type of thing?

Thanks,

-Tim

[Ricardo Niederberger Cabral]

Yes, that's the first test I'd try. 

Ideally it should not be necessary to add images to isk-daemon just for calculating their visual similarity value. I've created an issue for that:

https://github.com/ricardocabral/iskdaemon/issues/50

Best regards,

Ricardo Niederberger Cabral
http://www.imgseek.net/
ricardo.cabral at imgseek.net

-Tim

--

You received this message because you are subscribed to the Google Groups "imgseek_users" group.

To view this discussion on the web visit https://groups.google.com/d/msg/imgseek_users/-/qH_3Pbf-MB0J.