pica
is one more experiment with high speed javascript, from authors of
paco. It does high quality image resize
in browser as fast as possible.
When you resize images using modern browsers' canvas
low quality interpolation algorythms are used by default. That's why we wrote pica
.
- It's not as fast as canvas, but still reasonably fast. With Lanczos filter and
window=3
huge image resize (5000x3000px) takes ~0.5s on desktop and ~2s on mobile. - In modern browsers pica automatically uses Webworkers, to avoid interface freeze and use multiple CPU cores in parallel.
Why it's useful:
- reduces upload size for large images to pre-process in browser, saving time and bandwidth
- saves server resources on image processing
- HiDPI image technique for responsive and retina
- use single image for both thumbnail and detailed view
Pica is a low level library that does math with minimal wrappers. If you need to resize binary image, you should take care to load it into canvas first (and to save it back to blob). Here is a short list of problems you can face:
- Loading image:
- Due to JS security restrictions, you can load to canvas only images from same
domain or local files. If you load images from remote domain use proper
Access-Control-Allow-Origin
header. - iOS has resource limits for canvas size & image size. Look here for details and possible solutions.
- If you plan to show images on screen after load, you must parse
exif
header to get proper orientation. Images can be rotated.
- Due to JS security restrictions, you can load to canvas only images from same
domain or local files. If you load images from remote domain use proper
- Saving image:
- Some ancient browsers do not support
.toBlob()
method, use https://github.com/blueimp/JavaScript-Canvas-to-Blob if needed. - It's a good idea to keep
exif
data, to avoid palette & rotation info loss. The most simple way is to cut original header and glue it to resized result. Look here for examples.
- Some ancient browsers do not support
- Quality
- JS canvas does not support access to info about gamma correction. Bitmaps have 8 bits per channel. That causes some quality loss, because with gamma correction precision could be 12 bits per channel.
- Precision loss will not be noticeable for ordinary images like kittens, selfies and so on. But we don't recommend this library for resizing professional quality images.
node.js (to develop, build via browserify and so on):
npm install pica
bower:
bower install pica
Resize image from one canvas (or image) to another. Sizes are taken from source and destination objects. Return task ID to be able terminate it early.
- from - source canvas or image.
- to - destination canvas.
- options - quality (number) or object:
- quality - 0..3. Default =
3
(lanczos, win=3). - alpha - use alpha channel. Default =
false
. - unsharpAmount - >=0, in percents. Default =
0
(off). Usually between 50 to 100 is good. - unsharpRadius - 0.5..2.0. Radius of Gaussian blur. If it is less than 0.5, Unsharp Mask is off. Big values are clamped to 2.0.
- unsharpThreshold - 0..255. Default =
0
. Threshold for applying unsharp mask.
- quality - 0..3. Default =
- callback(err) - function to call after resize complete:
- err - error if happened
(!) If you need to process multiple images, do it sequentially to optimize CPU & memory use. Pica already knows how to use multiple cores (if browser allows).
Terminate resizing task by id, returned from resize function.
Supplementary method, not recommended for direct use. Resize Uint8Array with raw RGBA bitmap (don't confuse with jpeg / png / ... binaries). It does not use tiles & webworkers. Left for special cases when you really need to process raw binary data (for example, if you decode jpeg files "manually").
- options:
- src - Uint8Array with source data.
- width - src image width.
- height - src image height.
- toWidth - output width.
- toHeigh - output height.
- quality - 0..3. Default =
3
(lanczos, win=3). - alpha - use alpha channel. Default =
false
. - unsharpAmount - >=0, in percents. Default =
0
(off). Usually between 50 to 100 is good. - unsharpRadius - 0.5..2.0. Radius of Gaussian blur. If it is less than 0.5, Unsharp Mask is off. Big values are clamped to 2.0.
- unsharpThreshold - 0..255. Default =
0
. Threshold for applying unsharp mask. - dest - Optional. Output buffer to write data (callback will return result buffer anyway).
- callback(err, output) - function to call after resize complete:
- err - error if happened.
- output - Uint8Array with resized RGBA image data.
true
if webworkers are supported.
You can use it for browser capabilities detection. Also, you can set it to
false
for debuging.
false
by default. Pica can use WebGL when available, if you enable
this feature. But current implementation is not complete and not recommended
for production. Result is noisy and code can crash.
Though, you can enable it to investigate technology, test on different devices and provide feedback.
pica.WEBGL = true;
pica.debug = console.log.bind(console);
Pica has presets, to adjust speed/quality ratio. Simply use quality
option
param:
- 0 - Box filter, window 0.5px
- 1 - Hamming filter, window 1.0px
- 2 - Lanczos filter, window 2.0px
- 3 - Lanczos filter, window 3.0px
In real world you will never need to change default (max) quality. All this variations were implemented to better understand resize math :)
Pica has built-in unsharp mask. Set unsharpAmount
to positive number to
activate the filter.
The parameters of it are similar to ones from Photoshop. We recommend to start
with unsharpAmount = 80
, unsharpRadius = 0.6
, unsharpThreshold = 2
.
There is a correspondence between UnsharpMask parameters in popular graphics
software.
We didn't have time to test all possible combinations, but in general:
- Top level API should work in all browsers, supporting canvas and typed arrays.
- Webworkers support is not required, but they will be used if available.
- If you plan to use only pure math core, then typed arrays support will be enougth.
Note. Though you can run this package on node.js
, browsers are the main
target platform. On server side we recommend to use
sharp for better speed.
You can find these links useful:
- discussions on stackoverflow: 1, 2, 3.
- chromium skia sources: image_operations.cc, convolver.cc.