Bug 50117

Summary:

pixel interpolation error

Product:

WebKit

Reporter:

jay <jay>

Component:

SVG

Assignee:

Nobody <webkit-unassigned>

Status:

RESOLVED WORKSFORME

Severity:

Normal

CC:

krit, mdelaney7, zimmermann

Priority:

Version:

528+ (Nightly build)

Hardware:

All

OS:

All

Attachments:

Description	Flags
screengrab 1	none
screengrab 2	none
screengrab 3	none
screengrab 4 showing opacity	none
script to produce banding effect (ie not by design)	none
uses percentage width and height	none
Firefox screenshot	none
shows relative merits	none

jay

Reported 2010-11-26 10:22:47 PST

somewhere the rendering of pixel level graphics is broken, banding is readily evident in the attached images and is limited to Safari-webkit and Chromium, not able to be reproduced using Opera nor Mozilla I have attached a testcase that uses lerp to interpolate colour values, and produce gradients across 2 dimensions, rather than a simple linear gradient. I have not been able to identify precise reasons, however in a broad range of situations, using a variety of methods, banding is always present.

Attachments
screengrab 1 (90.09 KB, image/tiff) 2010-11-26 10:24 PST, jay	no flags	Details
screengrab 2 (35.84 KB, image/tiff) 2010-11-26 10:24 PST, jay	no flags	Details
screengrab 3 (108.09 KB, image/tiff) 2010-11-26 10:25 PST, jay	no flags	Details
screengrab 4 showing opacity (35.90 KB, image/tiff) 2010-11-26 10:26 PST, jay	no flags	Details
script to produce banding effect (ie not by design) (1.26 KB, image/svg+xml) 2010-11-26 10:27 PST, jay	no flags	Details
uses percentage width and height (1.32 KB, image/svg+xml) 2010-11-26 10:45 PST, jay	no flags	Details
Firefox screenshot (115.26 KB, image/png) 2010-11-26 17:49 PST, Alexey Proskuryakov	no flags	Details
shows relative merits (59.83 KB, image/tiff) 2010-11-27 00:36 PST, jay	no flags	Details
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.

jay

Comment 1 2010-11-26 10:24:29 PST

Created attachment 74946 [details] screengrab 1

jay

Comment 2 2010-11-26 10:24:53 PST

Created attachment 74947 [details] screengrab 2

jay

Comment 3 2010-11-26 10:25:26 PST

Created attachment 74948 [details] screengrab 3

jay

Comment 4 2010-11-26 10:26:03 PST

Created attachment 74949 [details] screengrab 4 showing opacity

jay

Comment 5 2010-11-26 10:27:42 PST

Created attachment 74950 [details] script to produce banding effect (ie not by design)

jay

Comment 6 2010-11-26 10:31:36 PST

apologies, need to qualify testcase better.

jay

Comment 7 2010-11-26 10:45:42 PST

Created attachment 74951 [details] uses percentage width and height

jay

Comment 8 2010-11-26 10:48:40 PST

it is not that Opera and Mozilla cannot be made to produce banding, rather it is far more difficult. In the particular use-case, not given as large file, I have not yet found a way to reduce banding, which is not present at all in other browsers

Alexey Proskuryakov

Comment 9 2010-11-26 17:49:24 PST

Created attachment 74957 [details] Firefox screenshot FWIW, I'm seeing the same in Firefox 3.6 on Mac.

jay

Comment 10 2010-11-27 00:36:55 PST

Created attachment 74959 [details] shows relative merits

jay

Comment 11 2010-11-27 00:41:02 PST

ap, please compare safarti, mozilla and opera in similar size windows etc. I have attached an example, not contrived,just as quickly as possible. full screen height then shifted left to overlay. there is a considerable difference. could raise issue with working group? obviously no client should interpolate white lines, where data has been provided. on the other hand there is a 'speed' issue, as interpolation takes cycles.

Dirk Schulze

Comment 12 2010-11-27 01:18:52 PST

(In reply to comment #11) > ap, > > please compare safarti, mozilla and opera in similar size windows etc. > > I have attached an example, not contrived,just as quickly as possible. > > full screen height then shifted left to overlay. > > there is a considerable difference. > > could raise issue with working group? > > obviously no client should interpolate white lines, where data has been provided. > > on the other hand there is a 'speed' issue, as interpolation takes cycles. Can you check with webkit trunk please? I can confirm a visual difference between Safari 5 and FF as well as Opera. But it seems to work in trunk. Just download a WebKit nightly.

jay

Comment 13 2010-11-27 01:46:59 PST

Dirk, using r72487

jay

Comment 14 2010-11-27 01:47:36 PST

Version 5.0.3 (6533.19.4, r72487)

jay

Comment 15 2010-11-27 02:14:27 PST

filed as errata: define pixel interpolation when scaling http://www.w3.org/Bugs/Public/show_bug.cgi?id=11417

Note You need to log in before you can comment on or make changes to this bug.