I’m still experimenting with Minos. If I hover with the mouse over instances of a Minos training set in the Teach Bench, I can see the correlation value for that instance. From what I understand, this is also calculated when adding new instances to a training set and goes along with the Correlation Threshold in the Training Parameters.
How can I calculate this correlation value for an instance programmatically?
The Correlation in conjunction with Minos is the result of a correlation of the instance image with the model image of a given Minos training set (MTS).
To get the instance, use MTSModelInstance() and then GetImageFromInstance() to get the image of the instance. To get the image of the model first use MTSModel() and then GetImageFromModel() to get the model image. Last use the OptCorrelation() with both the instance image (as parameter IMG) and the model image (as parameter TPL) to calculate the correlation.
Ah yes, I’ve stumbled across the OptCorrelation() earlier, but I was was not sure what the data type TPL was referring to. Thanks!
Can I use this method to correlate arbitrary images (other than from an MTS)? Are there certain requirements for the images to be met to calculate a correlation?
There are other means to correlate images (and locate correlation matches) in CVB, but you will need to have a Foundation license for these. For a detailed description check out the Correlations Functions section of the Foundation.dll in the CVB Manual.
TPL is a bit of a relic and I honestly don’t know why the signature wasn’t simply using IMG in the first place. I guess there must have been a reason for this, but this seems to be lost in the mist of time…
I’ll change TPL to IMG for the next release. For now it’ll need to suffice to say that TPL is just another (and by now superfluous) way of saying IMG…