Mathis/sharp
1
0
Fork 0
mirror of https://github.com/lovell/sharp.git synced 2025-12-19 07:15:08 +01:00
Code Issues Packages Projects Releases Wiki Activity

Switch default interpolator to bicubic #289

Browse Source 
Only use gaussian blur for non-linear interpolators

Improves performance of bilinear by ~15%

Add liborc to the packaged build to improve bicubic perf

Add examples of the various interpolation methods

Add bilinear vs bicubic to perf tests
This commit is contained in:
Lovell Fuller
2015-11-15 22:04:31 +00:00
parent 2678d761ba
commit 84fd1caa46
21 changed files with 935 additions and 442 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
docs/api.md
View File

@@ -201,14 +201,18 @@ Ignoring the aspect ratio of the input, stretch the image to the exact `width` a
Use the given interpolator for image resizing, where `interpolator` is an attribute of the `sharp.interpolator` Object e.g. `sharp.interpolator.bicubic`.
The default interpolator is `bicubic`, providing a general-purpose interpolator that is both fast and of good quality.
Possible interpolators, in order of performance, are:
* `nearest`: Use [nearest neighbour interpolation](http://en.wikipedia.org/wiki/Nearest-neighbor_interpolation), suitable for image enlargement only.
* `bilinear`: Use [bilinear interpolation](http://en.wikipedia.org/wiki/Bilinear_interpolation), the default and fastest image reduction interpolation.
* `bicubic`: Use [bicubic interpolation](http://en.wikipedia.org/wiki/Bicubic_interpolation), which typically reduces performance by 5%.
* `vertexSplitQuadraticBasisSpline`: Use [VSQBS interpolation](https://github.com/jcupitt/libvips/blob/master/libvips/resample/vsqbs.cpp#L48), which prevents "staircasing" and typically reduces performance by 5%.
* `locallyBoundedBicubic`: Use [LBB interpolation](https://github.com/jcupitt/libvips/blob/master/libvips/resample/lbb.cpp#L100), which prevents some "[acutance](http://en.wikipedia.org/wiki/Acutance)" and typically reduces performance by a factor of 2.
* `nohalo`: Use [Nohalo interpolation](http://eprints.soton.ac.uk/268086/), which prevents acutance and typically reduces performance by a factor of 3.
* `bilinear`: Use [bilinear interpolation](http://en.wikipedia.org/wiki/Bilinear_interpolation), faster than bicubic but with less smooth results.
* `vertexSplitQuadraticBasisSpline`: Use the smoother [VSQBS interpolation](https://github.com/jcupitt/libvips/blob/master/libvips/resample/vsqbs.cpp#L48) to prevent "staircasing" when enlarging.
* `bicubic`: Use [bicubic interpolation](http://en.wikipedia.org/wiki/Bicubic_interpolation) (the default).
* `locallyBoundedBicubic`: Use [LBB interpolation](https://github.com/jcupitt/libvips/blob/master/libvips/resample/lbb.cpp#L100), which prevents some "[acutance](http://en.wikipedia.org/wiki/Acutance)" but typically reduces performance by a factor of 2.
* `nohalo`: Use [Nohalo interpolation](http://eprints.soton.ac.uk/268086/), which prevents acutance but typically reduces performance by a factor of 3.
[Compare the output of these interpolators](https://github.com/lovell/sharp/tree/master/test/interpolators)
```javascript
sharp(inputBuffer)