Normalize feature vector for IMAQ custom classifier?

[Ard]

I wondered if it is necessary to normalize the features of the feature vector in the custom classifier that is included in the Vision toolkit. I could not find in the vision concepts manual if this is necessary. Normally a feature vector has to be normalised, because otherwise a feature with high values has more impact on the classification result than a feature with relatively small values. Does anybody have experience with this?

I am planning to have a custom feature vector to be classified with a nearest neighbour classifier from the IMAQ vision vi's.

[KevinC]

Ard,

I think you have a pretty good understanding of classification. You're right that a feature vector must be normalized to prevent large values from having a greater influence on results than small values. The NI Vision Custom Classifier does not require you to normalize your features, but for best results it would probably be wise to normalize each feature into the range 0 to 1.

Hope that helps,
Kevin C.
National Instruments
Vision Algorithms

[Ard]

Thanks Kevin,

I Think this helps me to better design the feature vectors and classification routines that I am designing at the moment. I've read through the vision concepts manual, but I couldn't find this topic of feature vector normalisation in the manual. Now it is clear to me how to do it the right way,

Ard Nieuwenhuizen

Wageningen University, Netherlands.