February 25, 2019

857 words 5 mins read

photopea/UPNG.js

photopea/UPNG.js

Fast and advanced PNG (APNG) decoder and encoder (lossy / lossless)

repo name photopea/UPNG.js
repo link https://github.com/photopea/UPNG.js
homepage
language JavaScript
size (curr.) 88 kB
stars (curr.) 1346
created 2017-03-28
license MIT License

UPNG.js

A small, fast and advanced PNG / APNG encoder and decoder. It is the main PNG engine for Photopea image editor.

Download and include the UPNG.js file in your code.

Encoder

UPNG.js supports APNG and the interface expects “frames”. Regular PNG is just a single-frame animation (single-item array).

UPNG.encode(imgs, w, h, cnum, [dels])

  • imgs: array of frames. A frame is an ArrayBuffer containing the pixel data (RGBA, 8 bits per channel)
  • w, h : width and height of the image
  • cnum: number of colors in the result; 0: all colors (lossless PNG)
  • dels: array of millisecond delays for each frame (only when 2 or more frames)
  • returns an ArrayBuffer with binary data of a PNG file

UPNG.js can do a lossy minification of PNG files, similar to TinyPNG and other tools. It performed quantization with k-means algorithm in the past, but now we use K-d trees.

Lossy compression is allowed by the last parameter cnum. Set it to zero for a lossless compression, or write the number of allowed colors in the image. Smaller values produce smaller files. Or just use 0 for lossless / 256 for lossy.

// Read RGBA from canvas and encode with UPNG
var dta = ctx.getImageData(0,0,200,300).data;  // ctx is Context2D of a Canvas
//  dta = new Uint8Array(200 * 300 * 4);       // or generate pixels manually
var png = UPNG.encode([dta.buffer], 200, 300, 0);   console.log(new Uint8Array(png));

UPNG.encodeLL(imgs, w, h, cc, ac, depth, [dels]) - low-level encode

  • imgs: array of frames. A frame is an ArrayBuffer containing the pixel data (corresponding to following parameters)
  • w, h : width and height of the image
  • cc, ac: number of color channels (1 or 3) and alpha channels (0 or 1)
  • depth: bit depth of pixel data (1, 2, 4, 8, 16)
  • dels: array of millisecond delays for each frame (only when 2 or more frames)
  • returns an ArrayBuffer with binary data of a PNG file

This function does not do any optimizations, it just stores what you give it. There are two cases when it is useful:

  • saving 16-bit colors (note, that PNG is big-endian, unlike Uint16Array in JS)
  • your image is too large, and “expanding” to 8-bit RGBA would use too much memory (e.g. 4-bit grayscale 50,000 x 50,000 = 1.25 GB, 8-bit RGBA would be 10 GB)

Decoder

Supports all color types (including Grayscale and Palettes), all channel depths (1, 2, 4, 8, 16), interlaced images etc. Opens PNGs which other libraries can not open (tested with PngSuite).

UPNG.decode(buffer)

  • buffer: ArrayBuffer containing the PNG file
  • returns an image object with following properties:
    • width: the width of the image
    • height: the height of the image
    • depth: number of bits per channel
    • ctype: color type of the file (Truecolor, Grayscale, Palette …)
    • frames: additional info about frames (frame delays etc.)
    • tabs: additional chunks of the PNG file
    • data: pixel data of the image

PNG files may have a various number of channels and a various color depth. The interpretation of data depends on the current color type and color depth (see the PNG specification).

UPNG.toRGBA8(img)

  • img: PNG image object (returned by UPNG.decode())
  • returns an array of frames. A frame is ArrayBuffer of the image in RGBA format, 8 bits per channel.

Example

var img  = UPNG.decode(buff);        // put ArrayBuffer of the PNG file into UPNG.decode
var rgba = UPNG.toRGBA8(img)[0];     // UPNG.toRGBA8 returns array of frames, size: width * height * 4 bytes.

PNG format uses the Inflate algorithm. Right now, UPNG.js calls Pako.js for the Inflate and Deflate method.

Quantizer

UPNG.js contains a very good Quantizer of 4-component 8-bit vectors (i.e. pixels). It can be used to generate nice color palettes (e.g. Photopea uses UPNG.js to make palettes for GIF images).

Quantization consists of two important steps: Finding a nice palette and Finding the closest color in the palette for each sample (non-trivial for large palettes). UPNG perfroms both steps.

var res  = UPNG.quantize(data, psize);
  • data: ArrayBuffer of samples (byte length is a multiple of four)
  • psize : Palette size (how many colors you want to have)

The result object “res” has following properties:

  • abuf: ArrayBuffer corresponding to data, where colors are remapped by a palette
  • inds: Uint8Array : the index of a color for each sample (only when psize<=256)
  • plte: Array : the Palette - a list of colors, plte[i].est.q and plte[i].est.rgba is the color value

FAQ

  • To get one common palette for multiple images (e.g. frames of the animation), concatenate them into one array data.
  • When working with less than four components, set the remaining components to a constant value (e.g. to zero)
  • When working with transparency, premultiply color components by transparency (otherwise, rgba(1,1,1,0) would be closer to rgba(1,1,1,1) than to rgba(0,0,0,0) - transparent mapped to white instead of transparent)
comments powered by Disqus