|
|
This is specifically for the case of premultiplyAlpha=false.
Following is an example. Given an image with only one pixel with value of
(255, 0, 0, 26), and we load the this image either as a Blob or a
HTMLImageElement and call createImageBitmap with premultiplyAlpha=false.
the ImageBitmap::copyBitmapData will return a pixel
value of (197, 0, 0, 26) which is totally wrong. The reason it fails
is that the SkImage that is extracted from ImageDecoder is always
in kPremul alphaType even if "AlphaNotPremultiplied" is given to
the ImageDecoder.
With this CL: https://codereview.chromium.org/1820733004/, the
extracted SkImage now has a kUnpremul alphaType. ImageBitmap
should adapt to this change as well. So this CL make correction
for the case of createImageBitmap from Blob and HTMLImageElement
so that copyBitmapData will succeed with the SkImage extracted
from ImageDecoder.
We have also updated corresponding layout tests. In particular, we change
an existing image's alpha channel from 128 to 26. In this case, if the
ImageBitmap constructor handles premultiplyAlpha=false by first premultiply
and then un-premultiply, we will see a large loss comparing with input.
Here we can see that the color is very close (tolerance of 2) to its input
image. Additionally, we added several new layout tests to test the
cases of createImageBitmap(ImageBitmap) where the parameter ImageBitmap
is created from various kinds of sources.
2 layout tests are deleted under fast/canvas/. The reason is that
these layout tests uses 2d canvas to display an ImageBitmap,
and it cannot distinguish an ImageBitmap created with premultiplyAlpha=
false with an ImageBitmap created with true. Now that we have
a large set of layout tests under fast/canvas/webgl, these 2 tests
are redundant.
Review URL: https://codereview.chromium.org/1809603004
Cr-Commit-Position: refs/heads/master@{#383004}
|
