Investigation of current color spaces for HDR content reproduction over the web
Presenter: Mekides Assefa Abebe (NTNU)
Duration: 10 min
Slides & video
Uh, my name is Mekides Assefa from uh, color lab in NITU Norway.
Today, I'm going to present to you on the topic of HDR image color fidelity on the world wide web.
The goal of color reproduction in color imaging is uh, reproducing real-world same, to human observers.
However, devices on different states of this pipeline have limited color reproduction capability.
Therefore, for good color fidelity, international color standards, and color management solutions are uh, necessary.
For color standards, uh, different international standards, the standardization organizations introduced various standards for different applications.
These standards define the system colorimetry or color space and transfer functions to transform different input signals to display devices.
To give examples of common ITU standards, the common BT.709, uh, which defines a high high-definition television, and the very recent BT.2100, uh, Rec 2100 which defines high dynamic range television contents.
In addition to having standard specification, color management is important for device interoperability with increased color fidelity.
Current color management systems are based on device independent paradigms.
Each device is color calibrated and only a transformation to a device independent colored space is needed to be determined.
Uh, too many examples of such uh, device independent paradigm color management systems are, uh the academic coloring coding system for motion picture image, ACES, and the ICC color profile based color management systems.
The color management system followed by uh most web browsers is uh, ICC color profile based.
Not all currently available browsers support color management.
Color management uh, systems supporting browsers currently include the Internet Explorer, Microsoft Edge, Firefox and Chrome for Windows, and Safari for, uh, OSX from Apple.
Color management systems supporting uh, browsers uh, considers embedded profile as a source, and an embedded profile into the image as a source and display profile as a destination profile and they apply transformation to and from the profile connection space.
Which is mostly considered to be XYZ color space.
The embedded and destination ICC profiles provide the color management model, the color space information, and other meta data and transformation matrix or uh, or lookup tables to and from the profile connection spaces.
And the color management module of the browser will take this meta data information and apply the transformations.
But most browsers assume sRGB as their standard coolor space, which means that uh every untagged image will be assumed as sRGB.
Uh, some browsers like Firefox, they also expand the image colors to the display gamut, unless their user specified it manually in the in the browser setting.
Uh, but currently the supported gamuts are sRGB and Adobe RGB.
HDR formats are not currently supported on browsers, there is currently a commendable work from uh, W3 uh, WorldWide Web consortium as a CSS Color HDR module level one for Cascading Stylesheets which recommends the application of HDR uh, supporting color spaces such as Rec.2100 with both uh, perceptual quantization and hybrid uh, log-gamma encodings.
Also the uh, JzAzBz and ICtCp kind of a perceptual uh, perceptually uniform color spaces.
Therefore I have simulated the upper the results of the image shown on the slide uh, on MatLab following the ICC color management process.
It can be seen that whatever color space we used as a profile connection space the results will be always darker.
For better rendering of HDR contents on SDR display, tone mapping is required.
A particular tone mapping can be included in the profile as a lookup table.
And I think a related work is currently taking place in ICC in a table to incorporate HDR contents in the new version of the new versions of ICC profiles.
Then the question will be what type of tone mapping is good for improving color color fidelity.
Tone mapping has been a wide field of research over the years.
Currently many tone mapping operators are available.
There are simple global tone mapping operators, based on gamma curve uh, based correction, or more perceptually uh, uh, designed sigmoidal curve based correction.
Sigmoidal operators uh, clip current detailed information from the image.
They introducing hue, uh they introduced hue and saturation changes, which are shown in the HSV color cloud visualizations so on the slide.
There are also local tone mapping operators, which preserve more details by supporting the reflectance and illumination information.
And then later compressing only the illumination information.
Such solutions are mostly computationally complex.
Therefore the choice of the based tone mapping operators that requires a compromise uh, to be done between the computational complexity and image quality or color uh, fidelity.
There are also more complex color appearance models designed with HDR tone mapping in mind.
The same tone mapping challenged such as over and underexposed of contents, hue, and saturation changes of colors, are also observed in these models.
The main source of the hue and saturation changes in my investigation are found to be the perceptual uh non-uniformity, and a strong interdependence of luminance and the chrominance channels of these uh perceptual color spaces.
Even so the ITU uh recommended uh ITU recommendation BT, BT.2390 a much simpler sigmoidal based tone mapping operators which will be competed depending on the black level and white levels of the targeted SDR displays.
So in this case, the idea is to keep as much details as possible in the mid-tone and uh clip as much in the shadows and highlights.
However, since all this is assumed that every HDR content have highlights and shadows, which doesn't contain as important details for the color appearance of the entire scene, which I find is not true all the time.
So far, I was only considering color uh managing HDR image for SDR displays.
Uh, but what if we have HDR displays on our tasks?
What will happen if the solution is uh if the situation is reversed?
Should we just expand uh everything to the display from capability uh to linearly just like uh the Firefox is uh Firefox browser is doing on uh white color gamut displays right now?
Or should we keep the embedded colors, the specifications in the image, and uh consider our expensive display as if it is just a CDR display?
These are also important question need to be answered by the browsers' color management modules.
I have investigated this issue in my PhD back in 2014 up to 2016 with uh SYNTH HDR display.
The hue shift and color desaturation problems were highly visible on HDR displays.
Also compared to the original HDR scenes we are actually expanse expanding the distorted HDR colors in case of reverse tone mapping operators.
Therefore, in addition to luminance expansion we found that uh saturation and color correction before the expansion, uh to be valuable for actually enhancing color fidelity.
So if you need more discussion on this topic you can find it on my PhD thesis.
The link is given in the final slides of this presentation.
Therefore in general for more accurate color reproduction, web browsers uh color management models need to incorporate perceptual and adaptive tone mapping, as well as gamut mapping transformations.
The, the operations need to consider gamut volumes instead of just the 2D gamut spaces.
So uh in summary, I have provided this um this points on the slide uh thirteen.
I have also provided uh references on the final slide
of number 14.
Thank you uh for your attention.