Closed dlebauer closed 6 years ago
The explanation is found here: https://support.bioconductor.org/p/90366/
Haralick features evaluates co-occurrences of gray value pairs of two adjacent image pixels and computes/derives probabilistic features from the co-occurrence matrix. As repeating texture pattern can have different scale, the "adjacent" pixel can be >1 pixel away. By downscaling the image it is easier and faster to compute the co-occ matrix, vs using a co-occ search window with <x,y> offset parameters.
Also a good reference is this: http://earlglynn.github.io/RNotes/package/EBImage/Haralick-Textural-Features.html
Hope this helps
Which scales are useful? Is 1 the appropriate scale?
That depends on the image and the texture property.
As for now I suggest to use 1 and perform a data analysis.
-- Kevin Nagel, Ph.D. Bioinformatician
On Fri, Sep 29, 2017 at 6:28 PM, David LeBauer notifications@github.com wrote:
Which scales are useful? Is 1 the appropriate scale?
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/terraref/extractors-stereo-rgb/pull/15#issuecomment-333173794, or mute the thread https://github.com/notifications/unsubscribe-auth/AGjwj2EZmuXuX1k0C4OHNjvstqFhLnvWks5snRqngaJpZM4PhGOQ .
Following discussion in https://github.com/terraref/reference-data/issues/172
Question: There is a an option to compute multiple 'scales' of Haralick metrics. The default is to compute scales 1 and 2, but can apparently be any integer vector. I've hard-coded this to compute only scale 1. What is a scale in this context, and is 1 the appropriate scale to compute by default? (this ? cross posted in https://github.com/aoles/EBImage/issues/24)
(note: reverts terraref/extractors-stereo-rgb#14 that reverted terraref/extractors-stereo-rgb#13, which was prematurely merged)