You can also do this via ops. A connected component on the binary image gives you a labeling which allows you to get the regions. Regions can then be used standalone to calculate geometric features of objects or in combination with the source image to calculate features based on the intensity values. Actually, some features are implemented n-dimensional, so you can use them for 1d, 2d or 3d.
We (=@tibuch) are currently going over the implementations again, validate them one more time and also implement pixel based geometric features which not require a
I think we could implement a way faster method for CoG, i.e. overload the existing one by one which is simply multi-threaded. Also the
Op matching in this case may slow down the computation a bit.