imagery-d v0.3.0

imagery-d v0.3.0

This release refines the public image-processing surface for consumers.

Highlights

• The default import imagery; facade exposes conventional image-operation
  names such as gaussianBlur, medianBlur, laplacian, canny,
  opening, closing and normalizedDifference.
• Parallel, partitioned and intermediate Canny stages remain available for
  advanced consumers through imagery.analysis.primitives without enlarging
  the default facade.
• Research notes and detailed benchmark evidence are maintained separately in
  imagery-d-research,
  keeping consumer source archives compact.

Compatibility

The package remains pre-1.0. Direct imports of the primitive module retain
compatibility aliases for medianFilter, laplace, open and close.

imagery-d v0.2.0

imagery-d v0.2.0 — RGB analysis and luminance stretch

This source release extends v0.1.0 Levels with RGB8 image analysis and
tone-mapping helpers. It remains a pre-1.0 D library for already decoded
rasters, with explicit pixel, ROI, mask and output contracts.

Added

• RgbHistogram256 and tryRgbHistogram256 count stored RGB bytes over an
  ROI and report per-channel minimum, maximum and mean. An optional binary
  mask selects samples; zero excludes, every nonzero byte includes once.
• Derived inclusive cumulative counts and integer nearest-rank percentiles
  require no second image traversal.
• Percentile contrast-stretch and histogram-equalization LUTs produce exact
  3×256-byte maps. The caller decides where to apply a map. A histogram mask
  selects samples for estimating the map, not pixels to edit.
• tryRgbLuminanceStretchOwned estimates bounds from XYZ-D65-Y luminance,
  scales linear-light RGB uniformly and publishes a new ROI-local owned
  normalized encoded-sRGB float image. Its optional binary mask estimates
  thresholds only; clipping policy and worker count are explicit.

Scalar Levels remains available with the v0.1.0 source contract. The
package now depends on raster-d ~>0.5.5 and color-d ~>0.3.0.

Compatibility and distribution

Linux x86-64 is the supported baseline. macOS and Windows remain best
effort. Pre-1.0 source APIs may evolve; no stable ABI or portable latency
guarantee is promised. The luminance operation is sequential by default;
applications can choose an explicit worker budget for previews.

This release is distributed as a GitHub source archive with API docs.
imagery-d is not registered on DUB; registration is planned for v1.0.0
unless a concrete earlier need arises.

imagery-d v0.1.0

imagery-d v0.1.0 — First release

The first pre-1.0 release of imagery-d, a general-purpose image
processing library for the D programming language. This is a small
usable starting point, not a feature-complete image editor or a
v1.0 stability promise.

Included

• Scalar Levels for already decoded raster-d views with
  ubyte, ushort or float samples.
• Per-channel selection, explicit ROI and optional 8-bit weighted
  editing mask; no implicit sRGB/ICC conversion or alpha policy.
• A fresh ROI-local, interleaved owned RasterLease!float result,
  with explicit byte budget for the new pixel payload.
• Two callable @safe tryLevelsOwned overloads; predictable normal
  returned-error behavior and retained prior output on failure.
• Ddoc and a consumer tutorial.

The package is MIT-licensed and depends on raster-d ~>0.2.0.

Compatibility

Supported: Linux x86-64; tested on Ubuntu 24.04 with
DMD 2.111.0/2.113.0 and LDC 1.41.0/1.43.0.

Best effort: macOS 15 and Windows 2025 passed ordinary DMD/LDC
consumer canaries; the full release matrix was not required there.

Pre-1.0: Source API may evolve in future 0.x minor versions;
no stable ABI, universal numerical ULP bound, process RSS cap,
image decoding, file-writing or automatic colour-management guarantee.

Basic use

import imagery : Levels, LevelsError, tryLevelsOwned;
import raster : RasterLease, RasterView, Region2D;

@safe LevelsError applyLevels(
    scope RasterView!ubyte image,
    ref RasterLease!float result,
    size_t maxResultBytes)
{
    bool[3] channels = [true, true, false];
    return tryLevelsOwned!ubyte(
        image,
        Region2D(0, 0, image.width, image.height),
        channels[],
        Levels(0, 1, 2, 0, 1),
        maxResultBytes,
        result);
}

Only LevelsError.none means success. The result owner is retained
on ordinary returned errors. The budget covers only the new float pixel
payload; a runtime OOM Error is not guaranteed recoverable.

Installation

Source: https://github.com/alex-1974/imagery-d

After the repository has been registered by its maintainer at
https://code.dlang.org/, the registry discovers SemVer tags.
Then DUB consumers can depend on imagery-d ~>0.1.0.

Registration and automatic registry indexing are separate from creating
this GitHub Release. Do not treat the registry as live until the
published imagery-d@0.1.0 can actually be fetched and consumed.

Qualification

On the last reviewed develop commit before release promotion,
DMD/LDC package tests, positive/negative API contracts, ordinary and
out-of-tree source-archive DUB consumers, 624-sample independent
90-digit Decimal oracle, P6 end-to-end image consumer and rendered
DDox checks passed. Linux compiler floor and clean
raster-d@0.2.0 registry-resolution tests passed. Performance
qualification was bounded to the demonstrated gamma-2 path.

Release publication requires checks on the exact release PR head
and actual tagged main commit. No new research or C++ parity benchmark
is necessary for this first 0.1.0 release.