Skip to content

v3.2.1: faster resizing, blur and GIF decoding

Choose a tag to compare

View all tags
@matmen matmen released this 16 Oct 09:53
Immutable release. Only release title and notes can be modified.
v3.2.1
80100ca

This release makes resizing, blurring and GIF decoding faster, and fixes a bug when enlarging with RESIZE_AREA. Nothing changes in the API.

Performance

1920×1080 photo 3.2.0 3.2.1
resize() to half (bilinear) 49 ms 15 ms
resize() to half (area) 33 ms 9 ms
resize() to double (bilinear) 392 ms 89 ms
blur(5) 205 ms 86 ms
decoding a 101-frame GIF 134 ms 83 ms

Resizing (#68)

  • The arithmetic is integer and the vertical pass runs first.
  • For opaque images, two color channels share one multiplication.
  • Bicubic resizing and images with alpha get 20–40% faster.

Blur (#69)

The three box blurs work directly on the pixels as 32-bit words, two channels per sum. Transparent images are blurred premultiplied in 8 bits, as browsers do.

Smaller speedups

  • GIF decoding (#70): frames are composited from their color indices instead of an RGBA copy of each, and they're no longer copied a second time. The output is bit-identical.
  • clone(): about 50% faster.

Fixes

  • Enlarging with RESIZE_AREA left transparent rows and columns where a pixel's center fell exactly on the edge of the filter (#68).

Upgrading

No code changes are needed. If you compare resized or blurred images against stored reference images pixel by pixel, expect small differences:

  • most values differ by at most 1 per channel (blur: 2 for radii above 10)
  • larger differences are limited to the color of pixels that are almost fully transparent, which doesn't show once they're composited
Assets 3
Loading