EigenFrame.PreProcessing 0.4.34

EigenFrame.PreProcessing

Pure, dependency-free astrophotography preprocessing algorithms for .NET — frame classification, sample decode, tile-pyramid generation, image statistics, tile-based calibration, and stacking/integration math. Written in F#, usable from any .NET language.

The library is format-agnostic: it operates on already-decoded sample buffers and metadata, never on files. You read your XISF/FITS bytes however you like and hand the library canonical buffers; it does the math. This is what lets the same algorithms run identically in a cloud service and a local desktop app.

dotnet add package EigenFrame.PreProcessing

<PackageReference Include="EigenFrame.PreProcessing" Version="0.2.0" />

Targets .NET 10. Floors FSharp.Core at 10.0.100 (deliberately low — any net10 SDK satisfies it). MIT licensed.

The one thing to know: the canonical contract

Every pixel function works in one processing format: single-channel / planar float32 (or float), range [0, 65535] — not [0, 1]. You convert your source samples into that range; the library never guesses. The per-source-format scaling is provided for you via SampleFormat:

type SampleFormat = UInt8 | UInt16 | UInt32 | Float32 | Float64

SampleFormat is the library's own enum on purpose — it does not reference XisfLib.Core or any image-format library. If your data comes from XISF, map its XisfSampleFormat to SampleFormat once at your boundary (a ~5-line match).

A few invariants worth honoring:

• Pure & synchronous. No I/O, no async — you own all reads/writes and thread the compute off your request threads.
• You own the buffers. In-place variants (e.g. subtractBiasInPlace, normalizeInPlace) mutate the slice you pass and are named accordingly. (Some functions still allocate-and-return while the API settles pre-1.0.)
• Tile geometry is library-canonical. Tile sizes, pyramid structure, and the [0,1] tile wire format are decided by the library so independently-built tiles line up — don't reinvent them host-side.

Pre-1.0: the surface may change between minor versions until 1.0. Pin a version.

Modules

Examples

Classify a frame from its FITS signals — the IMAGETYP value (or None if absent) and whether an NCOMBINE keyword is present (the master marker):

open EigenFrame.PreProcessing

match Classification.classify (Some "Light Frame") false with
| Classification.Classified (cat, sub) -> printfn "%A %A" cat sub   // Frame Light
| Classification.Unknown                -> printfn "unrecognised"

// Canonical persisted names, e.g. for a catalog row:
let category, subtype = Classification.names (Classification.classify (Some "Master Dark") true)
// "Master", "Dark"

Decode + measure a frame you've already read into a ReadOnlyMemory<byte>:

open EigenFrame.PreProcessing

let fmt = SampleFormat.UInt16                       // map from your source format
let stats = Statistics.computeStats pixelBytes pixelCount fmt
printfn "mean=%f median=%f" stats.Mean stats.Median // both in [0, 65535]

Build the L4 tile pyramid straight from source bytes:

let tileSize = Tiling.validateTileSize 512
for (tx, ty, brotliBytes) in Tiling.generateL4Tiles pixelBytes width height fmt tileSize do
    // write brotliBytes to your tile store at (tx, ty)
    ()

License

MIT.