460269 - Wrong color profile used for LG Flatron L1919S monitor

Status: NEW

(Core :: Graphics: Color Management, defect)

Description

[Samuel LB]

User-Agent:       Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1b1) Gecko/20081007 Firefox/3.1b1
Build Identifier: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1b1) Gecko/20081007 Firefox/3.1b1

If the monitor colorspace is sRGB (or unknown), then an image tagged as sRGB should not be transformed in any way. But Firefox performs some conversion anyway.

This happens In Firefox 3.1 beta 1. It happens on Linux only (tested on Ubuntu 8.04 x86), the Windows version works correctly (tested on WinXP SP2).



Reproducible: Always

Steps to Reproduce:
1. Open the URL and scroll down to the section named "ANOTHER PROBLEM"
2. Move the mouse over the logo.

Actual Results:  
The colors are different.

Expected Results:  
The untagged and sRGB-tagged logos should look identical.

All PNG images created with the GIMP image editor are tagged as sRGB, as well as images on mozilla.com/org. So this bug is very visible to the user.

Comment 1

[Samuel LB]

This is a color management bug so I think it blocks bug 455056 and bug 455077.

Comment 2

[Samuel LB]

I downloaded the sRGB_IEC61966-2-1_withBPC.icc color profile from color.org, and if I change the "gfx.color_management.display_profile" preference to use that file then colors look correct again.

So I guess the color profile is detected incorrectly (I'm not using any color profile so it should default to sRGB)...

Comment 3

[Samuel LB]

This bug appears to be screen/graphics card specific, because I can reproduce it on a clean installation [a Live CD] on one computer but not on another computer. My screen is a LG Flatron L1919S, and my graphics card is a ATI Radeon 9500 Pro.

This makes me believe the bug might not be in Firefox but somewhere else. So how do you determine the color profile to use on Linux?

Comment 4

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Somehow related or a dupe of bug 450923?

Comment 5

[Samuel LB]

I don't think so since the colors are correct on Windows (but I have XP not Vista) and the symptoms are slightly different: For me the colors are less colorful and light shades of yellow look green, the other bug seems to be about white color becoming light yellow.

Comment 6

[Bobby Holley (:bholley)]

Hrm. 

The fact that transforms are not optimized away if embedded profiles are srgb is a known issue (see bug 449826). Given that the transform precision isn't infinite, there can sometimes be subtle differences, even if the transform should be identity.

It seems weird to me that this is happening on linux only. Are you setting a manual .icc profile in about:config on both OSes, and only seeing the issue on linux? Assuming all machines are x86, the only platform-dependent part of things is when the OS is queried for the native system profile, but that gets overridden if you specify a filepath in about:config.

Comment 7

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Ok, I can see a similar thing and had it already commented to bug 455077 comment 3. Bobby, could this probably be the same issue?

Comment 8

[Samuel LB]

@Bobby Holley:

No, when I set the ICC profile manually everything works as it should.

Comment 9

[Bobby Holley (:bholley)]

@Samuel:

So the device space of the monitor can only be made known to firefox though the OS when firefox queries the OS for the system profile. So when you say the monitor is sRGB, do you know if you have your system configured to have an sRGB profile as the default?

grumble grumble LG monitors grumble

Comment 10

[Samuel LB]

I don't have any color profiles installed, so I guess sRGB is used then?

Comment 11

[Bobby Holley (:bholley)]

It depends. On linux the most likely action is for X11 to return NULL for the query and for mozilla to use the internal sRGB profile. However, it's also possible that your setup is querying the EDID of your LG monitor, and we've determined that the LG EDIDs are bogus.

Comment 12

[Samuel LB]

Yes, that's very likely to be the cause actually. The color profile on the driver CD of the monitor is quite different from sRGB, so the differences are very noticable in non-colorspace aware applications. So instead of using the ICM in a few applications I've reduced the blue component by 20% on the monitor. This gives a very good result IMO. But the EDID will of course contain LG's color profile.

I wonder if parsing the EDID is really a good idea, I don't think any other application does this and at least Firefox will look very *different* from other applications if it does. Wouldn't it be a better idea to just use sRGB and let the user change the color profile if he/she wants to (for example, with an extension)?

Comment 13

[Samuel LB]

Here's the EDID data for the monitor in case you need it:

00ffffffffffff001e6df24a97db0000
0a1001036a261e78eaec54a5584a9a26
215054a56b80314f454f614f81800101
010101010101302a009851002a403070
1300782d1100001e000000fd00384b1e
530e000a202020202020000000fc004c
31393139530a202020202020000000fc
000a202020202020202020202020007a

Comment 14

[Bobby Holley (:bholley)]

(In reply to comment #12)
> I wonder if parsing the EDID is really a good idea, I don't think any other
> application does this and at least Firefox will look very *different* from
> other applications if it does. Wouldn't it be a better idea to just use sRGB
> and let the user change the color profile if he/she wants to (for example, with
> an extension)?

Well, we're not parsing the EDID per se - we're querying the OS for the system-wide color profile, and the OS may or may not parse the EDID data. We would like to get the profile from the system, since that would allow firefox to blend seamlessly into any color-managed workflows. Right now a lot of the problems seem centered around LG, so we may have to do some kind of blacklisting operation and just use sRGB if an LG monitor is detected.

Comment 15

[Stewart Gordon]

Attachment: Screenshot of broken sRGB adjustment in Firefox

URL: http://www.libpng.org/pub/png/colorcube/colorcube-pngs-sRGB-CSS.html
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.1.2) Gecko/20090729 Firefox/3.5.2 (.NET CLR 3.5.30729)

This seems to be a case of the same bug occurring under Windows.  Notice that the RGB values actually rendered for the CSS backgrounds match those specified (thereby assuming that my display is sRGB), but the image colours are wrong.

Comment 16

[Stewart Gordon]

(In reply to comment #5)
> I don't think so since the colors are correct on Windows (but I 
> have XP not Vista)

Do you have JS and colour management enabled?

I'm using FF 3.5.2 under Vista.  The STR show the problem for me: the logo background is a slightly lighter shade of brown in the sRGB-labelled version.

> and the symptoms are slightly different: For me the colors are less 
> colorful and light shades of yellow look green, the other bug seems 
> to be about white color becoming light yellow.

Comment 17

[Stewart Gordon]

(In reply to comment #2)
> So I guess the color profile is detected incorrectly (I'm not using 
> any color profile so it should default to sRGB)...

Then why doesn't it detect the colour profile equally incorrectly for the CSS colours?

Comment 18

[Bobby Holley (:bholley)]

(In reply to comment #17)

> Then why doesn't it detect the colour profile equally incorrectly for the CSS
> colours?

CSS colors aren't color managed by default for performance reasons. See http://bholley.wordpress.com/2008/09/12/so-many-colors/ and http://muizelaar.blogspot.com/2009/06/qcms-color-management-for-web.html

Comment 19

[Stewart Gordon]

IOW, the out-of-the-box behaviour infringes the standards.  If nothing else, default behaviour should at least treat CSS colours and sRGB-tagged PNGs equally.

I've had a look at those pages, but I still don't see what these performance reasons might be.  After all, as the first of these begins to say, the average stylesheet contains far fewer colours than the average image.

Comment 20

[Jeff Muizelaar [:jrmuizel]]

It looks this bug is about misdetecting the color profile for the some monitors, when we fix bug 585074 it will make it easier to diagnose this.