Post by asgr on Apr 11, 2018 9:33:14 GMT 8
I stumbled across something that seems useful. Known as Haralick features, these are a set of image analysis diagnostics originally created for computer vision and used in machine learning etc. They are effectively summary statistics for the co-occurence matrix (which is very similar to a lag matrix). See wiki:
en.wikipedia.org/wiki/Co-occurrence_matrix
You can create them using EBImage by setting haralickstats=TRUE in ProFound. There is some computational cost to computing these outputs, so it is off by default. Creating them probably adds a few percent to your runtime. You get an additional haralick object in the list output which contains an Nseg [row] x 26 [column] data.frame containing all of the features discussed here:
earlglynn.github.io/RNotes/package/EBImage/Haralick-Textural-Features.html
If you are feeling adventurous you can check out the original paper too:
haralick.org/journals/TexturalFeatures.pdf
My hunch is they should prove useful for identifying particular classes of sources (e.g. stars and galaxies). Have fun experimenting, and comment below if you find some useful successes.
A
en.wikipedia.org/wiki/Co-occurrence_matrix
You can create them using EBImage by setting haralickstats=TRUE in ProFound. There is some computational cost to computing these outputs, so it is off by default. Creating them probably adds a few percent to your runtime. You get an additional haralick object in the list output which contains an Nseg [row] x 26 [column] data.frame containing all of the features discussed here:
earlglynn.github.io/RNotes/package/EBImage/Haralick-Textural-Features.html
If you are feeling adventurous you can check out the original paper too:
haralick.org/journals/TexturalFeatures.pdf
My hunch is they should prove useful for identifying particular classes of sources (e.g. stars and galaxies). Have fun experimenting, and comment below if you find some useful successes.
A