Image Processing for X-ray images

[muks]

The zip files contains good and bad image samples that i will need to segregate. I tried using a n-black and clustering threshold. Currently i acheived the segragation by using two seperate threshold for the outer dark chip and a seperate one for the outer cover . But i am not acheiving repeatability. Any suggestions?

[Andrey_Dmitriev]

I would like to suggest that you forgot to attach the images...

Andrey.

[muks]

ok and suggestion taken.

[rpursley8]

What is your criteria for good and bad?

[muks]

The placement of chip inside the capsule should be within acceptable limits. Roughly it should not hit the outer cover.

[muks]

Ok for those interested, I first determine the orientation and then used two thresholds and edge detection (for the non-curved side). Referencing to this side i find the internal chips orientation. The processing time is a bit on the higher side. Optimizing it...

[rpursley8]

Have you used the pattern matching functions in the Vision package?  If you have a template image that you can use, then you can create two pattern matching functions, one for the inner and one for the outer pieces (thresholding as you currently do for pre-processing).    Maybe then you can use net difference in offset and/or rotation to determine pass/fail.  The pattern matching vis seem to run pretty fast.

[rpursley8]

It seems to work.  Maybe not as fast as you want.

[muks]

---

rpursley8 wrote:
It seems to work.  Maybe not as fast as you want.

---

That was my first stop. To use two different templates but consistency was an issue when the cover and inner chip merged the match found was not proper. Thanks for trying.

[BruceAmmons]

When you created the template for the outer cover, did you tell it to ignore the inner contents of the case?  For the case, it should only search for the outer edge of the case, not the contents.

Likewise for the chip, don't include a lot outside the edge of the chip other than a few pixels to define the edge.

Seems like pattern matching would work very well in this case like rpursley8 says.

Bruce