search

imagej / imagej-ops

ImageJ Ops: "Write once, run anywhere" image processing
https://imagej.net/libs/imagej-ops
BSD 2-Clause "Simplified" License
88 stars 42 forks source link

Bugfix: Grey-value co-occurrence matrices should be symmetric #632

Open haesleinhuepf opened 4 years ago

haesleinhuepf commented 4 years ago

Hi all,

as discussed with a user and @imagejan on the forum we assume that the grey level co-occurrence matrices should be symmetric. Otherwise, the direction of the read-out would play a role in further post-processing of the matrix.

Maybe @dietzc has an opinion here.

Thanks!

Cheers, Robert

imagejan commented 4 years ago

I think we might want to have both a non-symmetric and a symmetric version of the co-occurrence matrix op. It seems for texture feature calculations really the non-symmetric matrix is used (although I find it surprising, because that means that mirroring an image results in different texture features...).

For reference, here's how it's done in scikit-image (where they default to symmetric=False and normed=False):

https://github.com/scikit-image/scikit-image/blob/2bf1a351a0c2e19fcd37168460a18954d4aabdd7/skimage/feature/texture.py#L15-L105