Add support for image sampling (with gradients) and image pyramids?

[ojwoodford]

I recently asked a question on the JuliaLang discourse site about what packages people use for creating image pyramids and getting image samples at non-integer locations (i.e. using an interpolation scheme). I didn't really get anything helpful back.

I couldn't see any support in Images.jl or in the broader JuliaImages collection for producing image pyramids. Is this correct? Given how important multi-resolution approaches are, this is surprising. Is this something people would like added? Any advice on where, and how?

I also couldn't see support for image interpolation (and sample gradients) within Images.jl, and an image doesn't seem to work with Interpolations.jl. Is this correct? Is this also something people would like added? Any advice on where, and how?

[mkitti]

See ImageTransformations.restrict. That will produce downsampled images in a systematic fashion. It is slightly different than the factor of 2 reductions used by other ecosystems though.

If these are especially large images, you might want to consider chunking with HDF5.jl or Zarr.jl.

ImageTransformations.warp is a slightly different function that does use Inteprolations.jl in its implementation.

We may want to consider how to encode this in a pyramid multiscale scheme such as that of OME-Zarr:
https://ngff.openmicroscopy.org/latest/#multiscale-md

[ojwoodford]

I feel like there should be an intermediate package, ImageInterpolations, that enables Interpolations on Images, that ImageTransformations then calls. Is this a good idea?