New Features and Improvements to AstroImages

[sefffal]

History: #29

Status: T-2 weeks until ready to review & merge

This PR is an overhaul of AstroImages to address the following issues:

• Integration with Images.jl: Integration with Images.jl #1
• Integration with Plots.jl: Integration with Plots.jl (or next-gen plotting package) #2
• Cannot plot large images: cannot plot large images #10

It implements a new AstroImage struct based on DimensionalData.jl that:

• Keeps track of mappings from array indices to parent indices during slicing (necessary for WCS.jl support)
• Stores FITS headers
• Stores a WCSTransform object to map pixel coordinates to physical coordinates and back. Object is generated lazily from FITS headers when needed.
• Thanks to DimensionalData, can apply reductions with named dimensions e.g. sum(img, dims=X).
• Broadcasting, transformations, etc

It implements a new imview function for rendering arrays to static images that supports:

• optional non-linear scales
• arbitrary colormaps
• clipping functions
• contrast & brightness adjustments
• safe handling of NaN, missing, +/- Inf pixels & empty arrays

It also implements plot recipes that:

• use image series instead of heatmaps and reuses imview
• render images with pixel-perfect locations matching DS9 in GR and PyPlot
• display custom colorbars that account for non-linear colorscales, etc.
• display WCS world coordinates ticks and arbitrarily warped grid lines
• set the axis and colorbar titles according to WCS and FITS headers
• displays any unseen dimensions as a title (e.g. when slicing cubes)
• visualize polarization data

Still TODO:

• update tests (Edit: started)
• Reproject integration
• recenter function for adjusting image origin
• Better interface for RGB composites
• Finish docs
• FileIO registration

Known issues:

• WCS gridlines & ticks broken on images that wrap around past 0
• ImageFiltering
• views currently copy headers instead of sharing them. Needs support from upstream DimensionalData

Gallery

imview(img; clims=percent(99), cmap=:seaborn_rocket_gradient)

implot(img, contrast=2, stretch=asinhstretch)

More:

[rafaqz]

Absolutely. We're discussing ways to implement a package like that over at: lorenzoh/Invariants.jl#2

[sefffal]

The docs aren't ready yet & there are still some rough edges but I think this is about ready to merge.

The last major outstanding thing is FileIO.jl registration.
I currently have

    # TODO: This should be registered correctly with FileIO
    del_format(format"FITS")
    add_format(format"FITS",
        # See https://www.loc.gov/preservation/digital/formats/fdd/fdd000317.shtml#sign
        [0x53,0x49,0x4d,0x50,0x4c,0x45,0x20,0x20,0x3d,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x20,0x54],
        [".fit", ".fits", ".fts", ".FIT", ".FITS", ".FTS", ".fit",],
        [:FITSIO => UUID("525bcba6-941b-5504-bd06-fd0dc1a4d2eb")],
        [:AstroImages => UUID("fe3fc30c-9b16-11e9-1c73-17dabf39f4ad")]
    )

in AstroImages.__init__() which is obviously a hack.

When running load("img.fits") FileIO goes down the list, so AstroImages will be used if installed and otherwise fallback to FITSIO.

Does anyone know the procedure to request that change in the FileIO registry? Do I need to run this the (other) FITSIO maintainers?

Edit: it would be nice for this to work with .fits.gz but I wasn't able to make that work. I'm not sure FileIO can differentiate .fits.gz from .gz.

[sefffal]

Even with the FileIO registration question outstanding, I think this is ready for a review and merge.
After merge, I can keep adding tests and continue progress on the docs.

[giordano]

I went through this very quickly. Amazing work, looking forward to future developments!

[sefffal]

Thanks @giordano and everyone else for your help and guidance along the way! Will merge now and prep to tag a new release.