go-images documentation

A pure-Go (no cgo) scikit-image-style image-processing library. Filters, edge detectors, morphology, geometry and colour transforms over an RGBA raster — no OpenCV, no libvips, no ImageMagick.

Image processing in most languages means binding a C library. go-images is a single portable module that cross-compiles to a static binary, accelerated by go-asmgen SIMD on the separable hot operations — so it beats scikit-image on box blur and Sobel. 100% coverage, validated against scikit-image.

import img "github.com/go-images/images"

src, err := img.Load("photo.png")
if err != nil {
    log.Fatal(err)
}
blurred, err := img.GaussianBlur(src, 1.0)
if err != nil {
    log.Fatal(err)
}
edges := img.SobelMag(blurred)
if err := img.Save("edges.png", edges); err != nil {
    log.Fatal(err)
}

API surface

Area	Functions
I/O	Load, Save, Decode, Encode, ToRGBA
Filters	GaussianBlur, BoxBlur, Median, Sharpen, UnsharpMask
Edges	Sobel, SobelX, SobelY, SobelMag, Prewitt, Scharr, Laplacian, Canny
Morphology	Erode, Dilate, Open, Close
Geometry	Resize, Rotate90/180/270, Crop, FlipHorizontal, FlipVertical
Colour	Grayscale, Invert, RGBToHSV, HSVToRGB, OtsuThreshold, Threshold, AdjustBrightness, AdjustContrast

Performance & architectures

The separable hot operations are go-asmgen SIMD-accelerated on amd64 (SSE2), arm64 (NEON) and s390x (vector); all six 64-bit targets run a portable scalar core. go-images beats scikit-image on box blur (1.8–2.6× single-thread, up to 10.3× on all cores) and now matches or beats it on Sobel (1.02–1.06× single-thread, up to 4.8× on all cores); the benchmark page is honest about where it does not.

Where to go next

• Roadmap — the plan and what ships today.
• Performance — honest benchmarks versus scikit-image.

Source: github.com/go-images/images · also the cgo-free image processor behind go-embedded-ruby's Image class.