Saturday, November 02, 2013

SMPTE Meeting on Wider Color Gamuts

AV Science publishes a report from SMPTE Annual Meeting mostly talking about UHD/4K issues: "many presenters agreed that increasing the resolution has less impact than other factors on the viewing experience. What's more important—what produces a real "wow" factor—is increasing the dynamic range, color gamut, color subsampling, color bit depth, and frame rate." The color gamut extension was the hot topics at the meeting. The current HDTV's Rec.709 gamut is extended to UHD's Rec.2020 one:

Some propose to extend it even farther to XYZ one to cover all visible colors and beyond: "the system would be entirely future-proof, accommodating any display technology that might be developed without having to create a new system all over again."

For image sensors, these wider gamuts are quite incompatible with small pixels due to a higher color crosstalk. If wider gamuts indeed manage to produce the "wow" factor, it could become a boost for a new generation of low-crosstalk pixels and color filter technology.


  1. What? The bottom diagram is just the normal tristimulus color space which requires imaginary primaries. Just because a color has coordinates in this space does not mean that it is physically producible. Real emitters can only have coordinates that fall inside the horseshoe curve. Any three real emitters will cover a triangle inside the horseshoe. To cover an area more comprehensive than a triangle requires more emitters. The people who make high-end printers understand this well.

    Be careful about talking about sensor crosstalk. It is used correctly to describe photons that pass through one filter and then are collected at a photodiode underneath another filter. It is incorrect to characterize overlap in the spectral bands of the various filters as crosstalk.

    1. True, the XYZ space requires more colors in the color filter. In the past Sony tried to use RGB+Emerald filters in its camera, and Sharp used RGB+Amber in its TV panels, but so far they failed to show a significant advantage.

      "It is used correctly to describe photons that pass through one filter and then are collected at a photodiode underneath another filter."

      This is what I meant. This crosstalk reduces the gamut that the sensor can reproduce.

      "It is incorrect to characterize overlap in the spectral bands of the various filters as crosstalk."

      I agree, assuming the filters are ideally shaped. In practice the shape is not ideal. The deviation from the ideal curve is a sort of crosstalk, although there is probably a better term for that.

    2. Really what is being described by non-ideal channel responses (not just the filter curves alone) is the residual after the signals are algebraically combined to match as closely as possible the tristimulus curves. To do this with three channels requires overlaps in the channel responses. Bayer filters do a pretty good job, prism separators generally overlap too little and the Foveon curves overlap too much.

      However, to provide more accurate color matching than is reasonably possible with three channels, five or six channels are needed. Sony and others have made such cameras and Philips at one time made some six-channel displays.

      The six-channel sets tend to include sources very near the red and violet end of the horseshoe curve, one near the place where the curve separates from the 45-degree line, one at the tip of the curve and then two somewhere along the left profile. If I remember correctly, this covers 98% of the visible color space.

      Sony has demonstrated how much difference this makes in their two-projector setup at there Tokyo demo facility. It (along with increasing the frame rate) is where development may be profitably invested now that it is possible to make home displays with stable color.

    3. @ Bayer filters do a pretty good job...

      It depends on the quality of Bayer filters. There is a lot of trade-offs in the filter design for small pixels: the pigment grains should be small enough, the peak transmittance should be high, etc. The filters response ends up being far from optimal from the gamut reproduction point of view, and there is a significant spectral crosstalk.

      Large pixels can have less trade-offs, but UHD sensors with large pixels are not cheap, not to talk about lens and the rest of the system.

    4. What D.G. described was optical crosstalk. Another kind of crosstalk is electrical crosstalk, e.g., photo carriers generated under green pixel may diffuse to red and blue pixels before collected. In this sense, the crosstalk caused by color filter may be correctly called spectral crosstalk.

  2. Vlad, as you stated here, crosstalk reduces the gamut that the sensor can reproduce. How then will increasing the display gamut "provide a boost for a new generation of low crosstalk devices"?
    Small pixels will still have the same reduced color gamut, and the same rendering or even more aggressive color rendering will be required to map colors from the device space onto the display color space. Please clarify how the widening of the color gamut of display devices minimizes the color crosstalk issue in small pixels.

    1. Well, now there might be somewhat less incentive to create wide gamut sensors, just because the display devices normally have sRGB/Rec.709 gamut. If the wider gamut become popular and provide the "wow" factor they talk about, the insentive would be higher. And there is a lot that can be done to expand the gamut, in both small and large pixels. Look at imec multi-spectral sensors, for example.


All comments are moderated to avoid spam and personal attacks.