Golden Template Matching - what exactly does the registration method do/does it work?

[b.ploetzeneder]

I've been playing around with Golden Template Matching, using VDM 2013.

There is one option "Registration Method" - you can select None/Perspective. The Vision Concepts Manual explains that this is supposed to add some registration /correlation method for more accurate alignment.

To get some ideas about the limits of this, I created a test image, and applied a pure mathematical perspective distortion, and tried it in VIsion Assistant and LabVIEW. It didn't seem to change anything, whether I used registration or not. I specifically look at the defect area and subtract it. Nothing. 

Any ideas what I am doing wrong? Does it need background to work? More texture?
Attached are the test images... , just plug them into VA and look for yourself. I can give you more realistic data via PM if necessary.

-bp

Edit: By shifting the alignment centers, I can actually get values that are a tiny bit different between those two methods. However, in my case it seems to go in the wrong way.. and I have absolutely no idea what is happening in the background. Anyone successful with this?

[KevinC]

There are two reasons that the experiment you are trying is not working well.

1. The image is not a good candidate for Golden Template Matching. It is a smooth gradient. Golden Template Matching is intended for high contrast images like text, logos, etc.

2. The image is too distorted. The registration is only intended to correct images that are at most a few pixels askew, typical of an industrial inspection application. Think of it as a small additional correction to the passed in alignment.

Hope that helps,

Kevin C.

National Instruments