APOD Color - Loss Of Information (RJN / JB Please Read)

[NoelC]

Hi guys,

Just wanted to mention that today's APOD image on the main page - "M83's Center from Refurbished Hubble" (http://antwrp.gsfc.nasa.gov/apod/ap091116.html) has been processed in such a way as to discard the color space information the Hubble Heritage Team built into the original file. This is what the image is intended to look like:



Specifically, because of a lack of color space information in the JPEG file, and the behavior of most browsers and image display programs, there has been an implicit assignment of the sRGB color space to an image that was published by the Hubble folks in the ProPhoto RGB color space, causing a significant shift in color and loss of vividness and richness.

I've seen this before in other APOD images as well.

Robert and/or Jerry, it appears you are using Adobe Photoshop ImageReady to prepare these images for web display. Whatever specific steps you're using, it appears that somehow you're both skipping the color space conversion AND the preservation of the color space information, thus the unexpected shift I noted above.

Since the Hubble folks choose to use ProPhoto RGB, may I respectfully suggest that the image preparation for the APOD page be done a little differently, converted to sRGB in a manner that preserves the original color intent. Here is what I recommend:

1. Set your working color space to sRGB, choose "Preserve Embedded Profiles", and check the "Ask When Opening" checkbox in the Photoshop Edit - Color Settings preferences.

2. When you open a file Photoshop will pop up a dialog when the file (e.g., the M83 image) has been published in other than your working color space (sRGB). I generally recommend converting to sRGB at this point, though you should be mindful that you could see loss of data due to clipping if the original image data has a very wide gamut. Please feel free to eMail me directly if you'd like to know more about how to avoid this.

3. Do your image manipulation to prep for the web page (e.g., resizing, cropping).

4. In Photoshop proper (not ImageReady) choose File - Save As, and save the image as a JPEG. This will save the image with the sRGB color space information intact.

The "Save for Web & Devices" function of Photoshop CS4 offers some alternatives to the above, but from what I can see in the JPEG files you don't appear to have CS4.

It's worth noting that at this point in time Internet Explorer is not color-managed (though Safari is). Your best bet is to assume all browsers and viewers can display sRGB.

As I mentioned above, please feel free to eMail me privately if you'd like more information or detail on the above.

-Noel
NCarboni@att.net

[Case]

This is what the image is intended to look like

IMHO, the color profile 'ProPhoto RGB' is just as arbitrary as 'sRGB IEC61966-2.1'. As HST never looked at terrestrial views, there is little reference frame for e.g. R=219, G=149, B=157 to look like a certain shade of red. Saying "it looks too red," or "not red enough" in color profile X, is only aesthetics. If using color profile 'ProPhoto RGB' makes it easier too see the scientific details that an astronomer is looking for, then that is a just as valid reason as applying 'sRGB IEC61966-2.1' to the raw RGB data for 'natural' color viewing.

(A graphic designer will use color profiles for color management, in order to get predictable colors on various output devices. That seems far away from any precise representation of Hubble colors.)

I agree though, that removing a (non-sRGB) color profile will result in an unwanted color shift as many will view it wrongly as sRGB.

[NoelC]

Assuming the Hubble people were looking at their screens when they prepared the image, I am making the assumption that "what the image is intended to look like" is what they were seeing when they all nodded their heads that the image was ready for publication.

I will agree that the image above probably actually looks less like what you'd see if you were out there viewing in visible light than does the image with the more muted colors. However, that's just coincidence.

And I'm not sure why they choose to use the ProPhoto RGB profile... I'm guessing someone told them it was "better". If one converts the raw data to "fit" within the sRGB color profile without clipping, which is easy to do with FITS Liberator, then there is no advantage to using ProPhoto RGB. Virtually the only thing using a non-sRGB working space seems good for is causing mixups like the one I'm pointing out here.

-Noel

[Chris Peterson]

Just wanted to mention that today's APOD image on the main page - "M83's Center from Refurbished Hubble" (http://antwrp.gsfc.nasa.gov/apod/ap091116.html) has been processed in such a way as to discard the color space information the Hubble Heritage Team built into the original file. This is what the image is intended to look like...

Your observations about converting color space are good ones (although it isn't so important anymore to actually convert to sRGB, as long as the profile is embedded in the JPEG; most browsers will now correctly interpret color profiles).

That said, IMO the cropped image used on the APOD main page has much better colors than the original image, which IMO is garishly oversaturated and doesn't come close to accurately representing the color of H-alpha emissions. The original image processors could take a few cues from the APOD version of the image (and they could avoid a lot of potential problems by releasing their images with sRGB profiles, not ProPhoto RGB; the former being standard for the Web, and the latter being poorly supported in general).

[NoelC]

I agree with you in general, Chris.

However...

most browsers will now correctly interpret color profiles

Neither Internet Explorer (by far the one most people use) nor Firefox properly interpret color profiles (the latter I just tested, and it's possible I missed a setting somewhere to turn color profile awareness on).

Which "most browsers" are you thinking of?

-Noel

[Chris Peterson]

Neither Internet Explorer (by far the one most people use) nor Firefox properly interpret color profiles (the latter I just tested, and it's possible I missed a setting somewhere to turn color profile awareness on).

Firefox absolutely handles color profiles correctly, since 2.5 I believe. Safari and Opera do as well; not sure about Chrome. My understanding was that IE8 finally added this feature, but I haven't tested it.

Of course, you do need to make sure that you actually embed the profile into the JPEG file, or it will be treated as sRGB.

[NoelC]

Hi Chris,

Internet Explorer ABSOLUTELY does NOT display images using color profile information. The assumption is as always; they're displayed using no profile, which amounts to sRGB. This is still "most browsers" in use on the planet by a longshot. My own recent actual browser statistics from a site I help manage:

Internet Explorer: 66%
Firefox: 22%
Chrome 5%
Safari: 6%
Opera: 1%


You say FireFox *is* doing color management properly. Fair enough. I found a reference that there is an "optional color management feature" that can be enabled. Edit: I have just installed Firefox 3.5, and it does appear to now by default honor the color space setting in the Hubble team's JPEG file. Thank goodness for progress, but we are not there until Microsoft gets on board.

Check your color management capabilities here: http://www.gballard.net/psd/go_live_pag ... files.html

This is an interesting table: http://diglloyd.com/diglloyd/free/WebBr ... l#Browsers

-Noel

[geckzilla]

If you could convince them to stop sampling up certain images when they deem them too small for today's usually high resolution viewers I'd be happy too.
I've noticed that they really try to leave the source unaltered, which you can view once you click the image to see the (usually bigger) version of the picture. So whatever processing gets done, at least there's always that.

[RJN]

Hi. I am just seeing this. Thanks everyone for your keen observations. I "prepared" that galaxy image. I do indeed, usually, use Adobe ImageReady to prepare images. It is old and reliable software that I have been using for a decade or so. It is no longer for sale individually. Whenever I get a new computer, I make sure it can run the ImageReady executable. What I really like about ImageReady is that you can blink compare reduced images with originals to see if a human (me) can easily discern a difference between a compressed file and the original. (One quickly learns that almost all web images are larger than need be in terms of human perception.) ImageReady also gives the ability to control size, compression fidelity, and reduce the colors in the color table.

Beyond that, I am a novice. I don't even know what sRGB is. I did notice that when processing this Hubble image, ImageReady changed the color table. This is the first time I have ever noted that ImageReady had done that. Like Chris, I thought the original Hubble image was prepared in color strangely, which is unusual for the Heritage team. Usually, the Hubble people are quite good. Oddly, ImageReady rendered it much better this time. I began to wonder if ImageReady fixed an unintended flaw in the image, and presented the way that the Heritage team really meant it to appear. Either way, I really liked the new image's color table, and judged its colors to be scientifically reasonable as well. (On occasions, I have changed other images, including false-coloring B&W images, sharpening images, and changing gamma.) I did none of these things here on purpose -- I just went with the ImageReady version. As usual, I linked to as high a version of the original image as possible without crashing browsers. I hope this is a help.

- RJN

[geckzilla]

Robert, ImageReady was kind of melded into the new Photoshop, I think when CS came out most of its functions went into the "Save For Web" option. So if you have a later version of Photoshop it's much better to use the "Save For Web" option in there and you'll find it quite familiar. If not, I guess it's mostly the same even since ten years ago (has it really been that long? good grief) ... There are a few new things but I can't remember when they got introduced. Mainly the ability to control what meta data is included, which can make a big difference in file size, itself.

As far as color profiles go, I discard them, especially for web work. I figure the vast majority of people are looking at old monitors with poor color calibration, anyhow. I've long since given up on any attempts to get colors to appear similar for everyone. That's the purpose of color profiles. Somewhat of a lost cause, if you ask me... but there are some who'd have my head for saying such a thing.

[Chris Peterson]

As far as color profiles go, I discard them, especially for web work.

Geez, don't do that. More people than ever are actually calibrating their monitors. There's no downside to embedding a color profile in a file for the Web- the added size is trivial. If people are using poorly calibrated monitors, it doesn't matter. But for the many people who actually take some care to get reasonable color, it makes all the difference.

[geckzilla]

You should see me try to match a hex color from my CSS to an image that still has a color profile on it, Chris. It doesn't work very well. Old habits die hard.

[NoelC]

Hi Robert,

Don't feel bad, there are VERY few people who really understand color spaces.

I do suggest upgrading your software though. As mentioned above the latest versions do still offer the same functionality of your older Imageready and more.

-Noel

[dlw]

I just found this thread and would like to make a serious suggestion for the APOD editors: Please -always- include an embedded color profile for all images, especially the 'full' or original versions. Even if some browsers don't yet support this, many do and those of us who save or print some of these images would benefit greatly from having an appropriate embedded profile. In my experience, -none- of the APOD images have an embedded color profile.

ICC color profiles have at least 2 main purposes: 1) to preserve 'true' color throughout an image processing work flow, and 2) to attempt to enable consistency in how an image is displayed on various devices, e.g., screens and/or printers. It is the latter that I am concerned about WRT APOD images.

Clearly a lot of astronomical images are "false color", i.e., not visible to the human optical system. Therefore 'true color' is less significant in such images. However, aesthetics aside, I assume the scientists who produced the original images saw them in a certain way, perhaps to bring out interesting detail, perhaps because they liked what they saw. We as APOD viewers would benefit from having a similar view, if for no other reason than to see what they saw.

To illustrate the need for embedded profiles, I have attached 3 versions of the same APOD image. The 'original' has no profile; the other 2 have potentially useful profiles. Of course, I can't predict how anyone else will see these images but on my screen they are quite different from each other.

I strongly recommend that the Adobe RGB_1998 ICC profile be embedded if no other profile is already in the image. It has a broad color range and avoids the potential problems that the ProPhoto profile can produce.

By embedding a profile, -any- profile, APOD viewers will be able to "see what the APOD editors see" across all platforms that are calibrated and support color profile transformations.

Thanks for your consideration of this issue and request.

[NoelC]

Your three images look identical; it's possible the forum software has converted the images into the sRGB color space. However, your point is valid - the main APOD publications don't seem to provide such interpretation even yet.

This page might be of interest to those wishing to see how well their internet browser works with images that have embedded document color profiles: http://ie.microsoft.com/testdrive/Graph ... fault.html

One thing to keep in mind: With Internet Explorer 9 about to release with color management built in (FINALLY!), that will mark the point where every major browser offers color management based on document profiles embedded in images. And it really works (I've been testing the betas). This will be a real milestone in Internet image display!

I strongly recommend that the Adobe RGB_1998 ICC profile be embedded if no other profile is already in the image.

I am starting to come to the conclusion that, given the growing availability of wide gamut monitors, this is the right suggestion. This will provide the best possible viewing experience to the people with the best equipment, and still not be THAT far off for folks with older systems without color managed software.

Some might say that sRGB would still be better because it will provide more consistent results for people who have not yet moved up to modern browsing technology, but given that APOD stays around for a long time I think planning for the future is the better approach.

RJN, if you are still in need of an explanation of color spaces and how color-managed software works, please feel free to contact me. It would be an honor to share my knowledge.

-Noel

[Chris Peterson]

Your three images look identical; it's possible the forum software has converted the images into the sRGB color space. However, your point is valid - the main APOD publications don't seem to provide such interpretation even yet.

On my system, the three images look very different. And I do mean very; nothing subtle at all.

If you're seeing them the same, it's because your browser doesn't handle color profiles correctly, or because you have some other software on your system that is interfering. The forum software is not altering the color profiles in the sources, however.

[owlice]

What Chris said -- these images also look different for me, under two different browsers. I actually prefer the third of the three images, though; the other appear over-saturated to me. The ProPhoto looks particularly bad.

[Chris Peterson]

What Chris said -- these images also look different for me, under two different browsers. I actually prefer the third of the three images, though; the other appear over-saturated to me. The ProPhoto looks particularly bad.

I also think the third image is the most aesthetic of the three.

As a rule, there are only two color profiles that most people use (if they use any)- Adobe RGB and sRGB. The two have similar gamuts (with the sRGB just a bit smaller). Especially for web astroimages, I can't see any reason to use anything but sRGB. That's what I always convert my images to before making web versions, and I always embed the sRGB profile as well (even though most display software will assume sRGB if no profile is embedded).

[rstevenson]

I was about to say I can see quite significant diferences Noel, but I wouldn't want to pile on. Perhaps if each of us says what we're viewing the images in/on Noel might be able to discover why he can't see the differences. I'm running OS X v10.6.6 on a 3-year old iMac, the latest version of Firefox. The images also look quite different in Safari but not at all in Opera or Google Chrome, so the browser is obviously making a difference.

Rob

[owlice]

MacBook Pro (a few months old) with:
Firefox 3.6.13 -- different
Safari -- different
Opera 9.64 -- no difference

Will report from a Windoze machine when I can.

[dlw]

Ideally, I think, the APOD image editor would preserve whatever profile came in the image or add one that appeared to preserve what the editor sees so that we see a similar image.

I agree about ProPhoto; I believe it was developed for a different use case and would not work well for APOD. Here is a web site that has a good discussion of ProPhoto http://www.luminous-landscape.com/tutor ... -rgb.shtml After reading that discussion, I believe the appropriate thing to do with an image that has ProPhoto embedded is to use PhotoShop to --convert-- it to Adobe RGB 1998, i.e., do not simply replace the profile (as I have done above and below) but convert the RGB data to display, as best it can, the same image but with a more commonly usable profile.

Of course, merely replacing a profile can sometimes bring out additional information. Please compare the images below. This technique might be chosen if the APOD editor believes it enhances the image's ability to illustrate the point being made in the 'explanation.'

DLW

[owlice]

I post a lot of images submitted to APOD on Asterisk; I often have to reduce the size of an image, but I would be loathe to do anything with color profiles (assuming I even could; I use Graphics Converter to manipulate images), because I would see trying to bring out whatever details as what the imager does. It's very rare that I do anything else with an image -- possibly crop it (maybe ... once? twice?).

Do imagers expect others to whom they send their images to decide on color balance?

[Chris Peterson]

I post a lot of images submitted to APOD on Asterisk; I often have to reduce the size of an image, but I would be loathe to do anything with color profiles (assuming I even could; I use Graphics Converter to manipulate images), because I would see trying to bring out whatever details as what the imager does. It's very rare that I do anything else with an image -- possibly crop it (maybe ... once? twice?).

Do imagers expect others to whom they send their images to decide on color balance?

Simply converting from one color profile to another should have only a minor impact on the appearance of an image. The extreme examples above were created by changing the assigned profile- essentially, the images are lying about their profiles.

Where you have a single forum, like APOD, I think the wise thing is to convert all images to sRGB. That will rarely introduce a noticeable color shift, but it will ensure the greatest possible consistency in the viewing experience. (If you're just posting an image for discussion purposes, it may not be worthwhile doing this. But for actual APOD images, or for images where color is specifically under discussion, conversion to sRGB makes good sense.)

[dlw]

I would be loathe to do anything with color profiles (assuming I even could; I use Graphics Converter to manipulate images), because I would see trying to bring out whatever details as what the imager does.

I completely agree. I was merely making the point that replacing is different than converting. A purpose of profiles is so the image can be converted, as necessary, to display on different output devices as close to the original intent as possible. When you look at an image on your computer screen, it has been converted from the RGB data in the image file to a different RGB color space calibrated to your screen such that you should perceive what the image originator saw.

The problem is what to do if there is no profile at all in the image. Adding sRGB or Adobe RGB 98 (which has a somewhat wider color range) would at least create consistency in what your readers see.

DLW

[RJN]

1. Set your working color space to sRGB, choose "Preserve Embedded Profiles", and check the "Ask When Opening" checkbox in the Photoshop Edit - Color Settings preferences.

It does not appear that my version of ImageReady (1.0, copyright 1998) has this option. If I could set this up to run automatically at no additional time cost to me, I would. But I am reluctant to buy and learn new software to fix a problem that is perceived by only a small fraction of APOD viewers. If you know how to adapt my "old reliable" software to do this, please do tell me and I will try. Beware that I am a seasoned curmudgeon who frequently considers the use of reliable older products a sign of efficiency. Therefore, if no easy fix is forthcoming, please do not be insulted if I stand pat for now.

- RJN