@jsquash/resize

An easy experience for resizing images in the browser or V8 environment. Powered by WebAssembly ⚡️ and Rust.

ℹ️ You will need to have an already decoded ImageData object to resize. This can be accomplished by using other jSquash modules or using the Canvas API, if available.

Uses squoosh's Resize module. Composed of:

• https://github.com/PistonDevelopers/resize
• https://github.com/CryZe/wasmboy-rs/tree/master/hqx

Addtionally we have added support for the Magic Kernel algorithm for resizing images. This is a Rust implementation of the algorithm.

• https://github.com/SevInf/magic-kernel-rust

A jSquash package. Codecs and supporting code derived from the Squoosh app.

Installation

npm install --save @jsquash/resize
# Or your favourite package manager alternative

Usage

Note: You will need to either manually include the wasm files from the codec directory or use a bundler like WebPack or Rollup to include them in your app/server.

resize(data: ImageData, options: ResizeOptions): Promise

Resizes an ImageData object to the specified dimensions.

data

Type: ImageData

options

Type: Partial<ResizeOptions> & { width: number, height: number }

The resize options for the output image. See default values.

• width (number) the width to resize the image to
• height (number) the height to resize the image to
• method? ('triangle' | 'catrom' | 'mitchell' | 'lanczos3' | 'hqx' | 'magicKernel' | 'magicKernelSharp2013' | 'magicKernelSharp2021') the algorithm used to resize the image. Defaults to lanczos3.
• fitMethod? ('stretch' | 'contain') whether the image is stretched to fit the dimensions or cropped. Defaults to stretch.
• premultiply? (boolean) Defaults to true. Filters RGB weighted by alpha and returns straight RGBA, preventing transparent colors from bleeding into visible edges. Applies to all Magic Kernel variants as well as the standard resize methods. Magic Kernel performs alpha conversion in floating point without modifying the input pixels; set false to filter its channels independently.
• linearRGB? (boolean) Defaults to true

Magic Kernel methods filter through a bounded row cache instead of converting the entire input to floating point. This supports large supersampled inputs without the multi-gigabyte floating-point allocation required by earlier versions. Pixels are still alpha-weighted in floating point, and sharpening uses the full image's coordinates so there are no tile seams.

Example

import { decode } from '@jsquash/jpeg';
import resize from '@jsquash/resize';

const imageBuffer = fetch('/images/example.jpg').then(res => res.arrayBuffer());
const originalImageData = await decode(imageBuffer);
const resizedImageData = await resize(originalImageData, { height: 300, width: 400 };

Manual WASM initialisation (not recommended)

In most situations there is no need to manually initialise the provided WebAssembly modules. The generated glue code takes care of this and supports most web bundlers.

One situation where this arises is when using the modules in Cloudflare Workers (See the README for more info).

The main module exports initHqx and initResize functions that can be used to manually load their respective wasm module.

import resize, { initResize } from '@jsquash/resize';

initResize(WASM_MODULE); // The `WASM_MODULE` variable will need to be sourced by yourself and passed as an ArrayBuffer.

resize(image, options);

// Optionally if you know you are using the hqx method or magicKernel method you can also initialise those modules
import { initHqx, initMagicKernel } from '@jsquash/resize';

initHqx(HQX_WASM_MODULE);
initMagicKernel(MAGIC_KERNEL_WASM_MODULE);