Thursday, December 02, 2010

Sony CineAlta Moves Beyond Bayer

Hugo Gaggion, Sony Broadcast & Professional Solutions CTO, presented 2K/4K roadmap at NYC 2010 HD World show in mid-October. Sony published a video of this presentation discussing color filter pattern innovations, among many other things.

The previous generation of Sony CineAlta CCD-based Super 35mm cameras used linear RGB pattern:

To maintain the high resolution in green in all directions, including diagonal one, Sony proposes a new Q67 pattern for new 4K camera systems:

Bayer Pattern can produce a lot of artifacts:

Sony Q67 pattern produces less color artifacts:

Eventually Sony plans to switch over to full-color 4K sensors:

The competitive landscape on Super 35mm camera market, as Sony sees it:

Thanks to A.A. for sending me the link!


  1. this proposed new Q67 pattern, its same as bayer but rotated 45deg.
    Thus, with alternative (read: more sophisticated) decoding you should get same result with existing pattern than Q67?

    what did I missed?

  2. I love the slight of hand where they double the number of photosites on their Q67 (Bayer rotated through 45º) compared to the normal Bayer pattern. A Bayer pattern CFA with the same number of photosites as their "Q67" and a good demosaic will actually produce a better, higher resolution image.

  3. get full rgb, high dpi display, and very high resolution. Its going to be eye candy :)

  4. No innovations at all!

    It only say more resolution needs more photosites.

    Q67 pattern is almost same as bayer.

  5. O.K. even if they are right about RED ONE when they are out with SONY 4K sensor RED will be out with 5K EPIC and SCARLET going to 27K EPIC in few years! So, SONY is again to little to late!

  6. Its says clearly that on the graph above that the Sony 4k camera will actually have a 6k sensor (6k x 3k).

    The problem with a bayer sensor is that after debayering you aren't getting the full effective resolution. Which is why the Epic is 5k and the Sony is 6k even when the target resolution is 4k.

    The point of Q67 is that it doesn't need debayering. In fact, its not a new concept.

  7. > "The problem with a bayer sensor is that after debayering you aren't getting the full effective resolution."

    I disagree. First, if the reason you consider Bayer to have lower resolution is in part due to the fact that it's typically used with stronger OLPF filtering than types that are less susceptible to chroma aliasing (such as 3-chip and other CFA), then I would point out that to me, luma aliasing is nearly as heinous a crime as chroma aliasing (e.g. color moire), and therefore I consider it just as necessary.

    Second, the luma resolution with a good demosaic algorithm is about as good as a monochrome sensor with the same pixel count and OLPF. It's only the reduced chroma resolution that can justifiably be criticized. But I consider that an advantage because the extra color resolution is only visible in the rarest of real-life circumstances, and at some point between capture, processing, and display, will almost certainly be thrown away through compression anyway. For example, in most programs, even the very highest JPEG quality setting uses 4:2:2 chroma subsampling.

    Between the decades of improvement in Bayer demosaic algorithms and the benefits of throwing away unimportant information (full chroma sampling) early on, I think Bayer is still the best solution.

  8. @ Daniel

    For the applications that these 4k video cameras are discussing, 4:2:2 is a massive hindrance. Most require 4:4:4 because a high dynamic range cinema gamma video (that these cameras will output) will produce a video sample that will require a significant amount of grading.

    During this process for a 4:2:2 source the chroma subsampling results in banding in highly saturated areas of the video.

    Which is in this article they are trying as hard as possible to regain RGB resolution.

  9. "For the applications that these 4k video cameras are discussing, 4:2:2 is a massive hindrance."

    First, even if I take your word for it, the very people who are designing these cameras don't seem to agree with you. John Galt is the Senior VP at Panavision that designed the Genesis (which is essentially identical to the Sony F35 discussed in this post). During a CML discussion about whether there is any benefit at all to 4:4:4, he said:

    "Based on extensive image composite testing done by myself and my engineering group...It is extremely difficult to demonstrate a improvement even in color difference blue screen/green screen photography (because of the equal full resolution of 4:4:4 RGB). The subject matter includes liquids, flame, smoke and graduated color on a fine mesh cloth, all designed to exercise the process to its limits..."

    If anyone would be biased to overstate the benefits of 4:4:4, it would be the one who builds and sells one of the very few CFA cameras that is actually capable of delivering it. Yet here he actively recommends 4:2:2 instead. He goes on to state:

    "The color difference compositing process for both film and electronic imaging was invented by Petro Vlahos, for which be has won two separate Academy wards, one for film and one for electronic compositing. Petro also founded the Ultimatte corporation. I think it is telling that Ultimatte's latest hardware compositing system no longer supports 4:4:4 but still enables beautiful composites of smoke, liquids and fire without artifacts."

    Maybe he didn't know of or think to mention the advantage you mentioned -- that 4:4:4 has in avoiding banding when post processing files with high dynamic range.

    Second, if we restrict the discussion only to those rare circumstances where 4:4:4 will be used *and* subjects that will actually benefit from high chroma resolution, I *still* maintain that Bayer is superior. Sony's disingenuous comparsion between 18 MP Q67 and 9 MP Bayer might fool some people, but in a level playing field, 18 MP Bayer with a good demosaic and downsample would provide just as much chroma resolution as the 18 MP Q67.

  10. Daniel, I have no doubt that with the proper software, it's possible to pull essentially as good a key from 4:2:2 as 4:4:4.

    But grading is a different story. As you start moving colors around and changing hue, 4:4:4 has a distinct advantage. Now perhaps this advantage could be minimized by clever software. But as it stands right now, 4:4:4 is more gradeable.

  11. Daniel: demosaicing can only do so much. Show me a demosaiced picture and a sharp full-resolution picture and I'll tell you which is which in a heartbeat. I agree that you might extract marginally more detail in a 18 MP Bayer image, but you're going to end up with twice as many pixels to process while you'll only have a few percent more detail. High-resolution high-framerate video is hard enough to process today that this tradeoff makes sense.

    Also, if your goal is to reach a desired resolution like 4K x 2K, a Q67 sensor of this resolution will certainly give much better quality than a 6K x 3K Bayer sensor that you would downsample to 4K x 2K.

    Anecdotically, if extracting every last bit of detail is important to you while performance is not very important then you could demosaic 18 MP Q67 to a 36 MP image with as much detail as a 18 MP Bayer image -- or you could turn your screen 45° and demosaic it as a plain 18 MP Bayer image :-)


  12. Who is John Galt? ;-)
    Bayer is very tried and true. In analyzing a number of options, it always seems that the noise component benefit from Bayer (low cross-components) outweighs other draw-backs. Of course, what is shown is just Bayer at 45º. This is 'technical marketing'.


All comments are moderated to avoid spam and personal attacks.