Detailed Panasonic Colour Splitter Sensor report.


Last week Panasonic announced a new sensor tech that basically signs their first step out form the classic Bayer sensor paradigm. Now you can read a more detailed report about the new Colour Splutter sensor tech at readcube (Click here). Panasonic writes:

Sigma was the first company getting rid of the color filters. And now it’s Panasonic turn. And plenty of patents from Fuji, Canon and Sony shows that they are also working into the same direction. Exciting time ahead on the sensor front!

Found via Image Sensor World.

  • Will

    Anything like the sigma sensors iq and it will be revolutionary, the micro contrast is insane.

    • CookieOfFortune

      You wouldn’t really get much increase in sharpness compared to a normal Bayer filter sensor (therefore it would still have less sharpness compared to the Foveon sensor), but you would get at least a stop of ISO performance.

      • Renato S.

        I think you would get an increase in sharpness, since because of the Bayer color filter array you end up using anti-aliasing filters, I think that this would not need these extra filters, therefore, it would increase sharpness.

        • Stupig

          This design still needs AA filter. That said, one can always do away with AA filter and pay the price of aliasing – how the aliasing affects the visual quality is totally another matter. For example, the X-Trans CFA layout has irregular aliasing and is usually visually unnoticeable.

          • Will

            The Foveon sensor is a strange one, it’s like going back about 5 generations but then forward a couple of steps in a different direction, it’s probably the most exciting thing out their for IQ nuts and most people don’t even realise it.

            I don’t know the equivalence of this new Pany sensor tech to it but I hope camera manufacturers start heading in that direction.

            • uth

              it’s different from Foveon :)

            • Esa Tuunanen

              Foveon is out of the frying pan into the fire solution.
              Fixing one thing and breaking another.

              “There’s also the issue of differential sensing and uncorrelated noise, making reading low values from the sensor twice as noisy as a non-differential Bayer, so it’s really more like 8x (or that the Foveon actually only uses about 4% of the light, in a low light situation”

              Hopefully Panasonic is researching this design as long term goal:
              Because that’s what the ideal sensor would be like: Full per pixel colour data like in Foveon but without totally messed up signals of stacked photosite design.

          • Boooe

            >X-Trans CFA layout has irregular aliasing and is usually visually unnoticeable.
            Bullshit. It is as “unnoticeable” as far as Bayer CFA go

        • EspenB

          You need a low pass filter in any sampled system to avoid aliasing. Remember the basic sampling theorem of f > 0.5 fs. For any transfered frequency over 0,5 fs there will be aliasing artifacts.

          • EspenB

            Sorry. f < 0.5 fs

  • Bob B.

    So…can someone explain this in English? I this a breakthrough or just more marketing baloney?
    I guess more will be revealed in time….

    • Breakthrough.
      Although it still cannot solve the necessity of AA filter, the light capture has almost 100% efficiency – the only sensor design to date which can do this, beside the monochrome one used in high-end Leica cameras.

      • No, partial breakthrough.

        The Panasonic design has the same problem that the old CYMG sensors had. CYMG was tried as a substitute for RGBG (Bayer) because CYM filters take away less light than R&B filters do. The hope was that by letting more light through you’d get better low light performance. To a large degree that was true, but the problem is that you need RGB pixels eventually, which means that you have to move the values in color space. That turned out to be the problem with CYMG and it will be a serious problem with W/W-R/W-B too.

        But the gain in the Panasonic approach is higher than in the CYMG approach. So the question is whether the problems with moving the colors around after the fact is less than the gain from extra light. It almost certainly is given Panasonic’s presented data, but is it at breakthrough levels? I doubt it.

        I’d also be a little concerned about telecentricity with this design. I suspect that you need higher telecentricity for it to work than you do with Bayer, and that puts some restrictions on lenses.

        • but the problem is that you need RGB pixels eventually, which means that you have to move the values in color space. That turned out to be the problem with CYMG and it will be a serious problem with W/W-R/W-B too.

          I cannot figure why this transformation would be a serious problem, moreover, when current sensor resolutions are usually much higher than the lenses should serve out. The transformation is a quite simple linear equation thus no information loss appears when it’s performed.

          • El Aura

            You are not mixing single wavelength colours but distributions around a peak colour. And the resulting calculated red pixel should have a spectral distribution somewhat similar to that of the red receptor in human eyes.

            • As the color resolution of the human eye is anyway just the fraction of its monochrome one I cannot see a big problem problem even if the colour reproduction suffers a bit. But – as seen above – the preliminary tests show a good color reproduction and much more gain in light.

              • Esa Tuunanen

                Problem isn’t in human eye’s colour resolution but the fact that just like Foveon this sensor doesn’t output RGB signal but mishmash with colours messed together.

                Extracting RGB data from that mess is going to add noise/lower real efficiency.
                Question is now how much more does removal of colour filter increase amount of light/photons hitting photosites than required processing lowers efficiency.
                Remember that also Foveon has 100% light capture efficiency (all light getting to photosites) but all that processing lowers final efficiency even far below Bayer whose colour filter array filters away 50% of green light and 75% of blue and red light.

                • Even legacy colour television systems used to work with a luminosity signal and two color difference signals – moreover, both having much lower bandwidth than the luminosity signal. And even so the quality was surprisingly better than one would think.

                  • NTSC was referred to by most engineers jokingly as meaning Not The Same Color twice.

                    Artifacts show up in different ways in transforms. As Esa points out the usual problem is that you end up with noise in the color channels. Remember, noise is the thing we’re trying to avoid by removing the Bayer filters, but we’re adding it back in via another method. That’s one reason why I wrote “partial breakthrough”: the implied gain was one stop (actually 0.85 stop IIRC from the original Panasonic documents), the actual gain will be less than that. Gains are gains, sure, but as Sony has proven with the EXMORs, reducing read noise brought a big bang for the buck, and has no downside. We used to have DSLR cameras with read noise in the 10 e- range. Now we have one that’s 1.4 e-. The signal to noise ratio is generally stated as saturation minus read noise. The smaller the sensor is, the lower the saturation number is, making it more important to lower read noise. But if you then add in a new computational noise, you don’t make a gain.

  • CookieOfFortune

    They talk about issues with color purity in their current version. Let’s see if their R&D efforts can alleviate that. It would be quite revolutionary if they could get competitive color purity.

  • taran

    Should bring panasonic up to the performance of Nikon D2h.

    • true homer

      hahahaha!!! it was a joke right?

      • taran

        Sadly, a D2hs will still handily beat a GF5 for high ISO, almost 10 years after the d2hs sensor came out. So yes, panasonic designing sensors is a joke.

        • OldAlaskan

          There’s no need to be sad! I suggest that you do a comparison for yourself using Imaging Resource’s Comparometer. The Nikon handles the reds better but in every other area the GF5 is far ahead of the old D2h. The D2h was good for its day but it sensor doesn’t come close to the GF5’s.

          • true homer

            Let alone a gh3 or omd…some people will post just any ol bullshit

  • Ross

    Admin. The typo ‘splutter’ actually sounds OK in this. ;) :D

  • The only drawback I can see in this technology is the massive computational overhead. Processing the image may be considerably slower and more battery-hungry than what can be achieved using a colour matrix. Foveon sensors suffer from similar issues.

    • Actually there needn’t be much overhead. They measure a linear combination of signals and one can use this to demix to RBG by a fancy version of what would otherwise be a 3×3 matrix inversion. This is all compiled into ASICs anyway (i.e. this is done in hardware), so the computational overhead is probably quite minimal. It’s different for algorithms that’ll be programmed in the Venus engine.

  • Narretz

    Experimental tech. But since they made it public, they seem to be confident, they can apply it to consumer products. We’ll wait.

    • > Since they made it public…

      In Japan, one of the speculations in the business press is that Panasonic wants to unload their sensor business. Puts a slightly different spin on why they might want to out an interesting new technology, doesn’t it?

  • Yun

    I’ll only buy a new panasonic camera with this sensor inside it . The ISO performance have to go up at least by 1.5 FStop too . That is what it should achieved .

    • true homer

      Otherwise youll stick with olympus with panasonic sensors? Or do we still think sony now somehow makes live mos sensors?

      • Ross

        I think we still think (or know) that Sony is making the 4/3’s sensor for Olympus.

        • Anonymous

          +1. Everyone in the photographic world except the Panasonic suicide bomber brigade.

          • true homer

            So sony now somehow can make live mos sensors without getting into legal problems with Panasonic? Live mos is licensed under Panasonic and they themselves said they havent sold rights to it. The comment that the omd sensor was sony was a bad translation that was taken down by the original source. But people will beoieve anything, people will actually believe first that panasonic used a sony sensor in the gh3 than one of their own…

  • spam

    Did Sigma/Foveon get rid of the color filters? I know they sample all colors in each pixel, supposedly by letting each layer “abosrb” only one color and letting the rest of the light thru. But do all the rest of the light really pass thru or? It certainly don’t look that way when you look at high ISO performance which should be excellent if all the photons are used, but actually is a lot worse than any other big sensor camera on the market today.

    As for Panasonic’s solution, it will need demosaic algorithms similar (but different layout of course) to what’s done with the Bayer sensor output. To me it seems like the algorithms will need to be more complex as they never know exactly how much extra light a pixel get from it’s neighbouring pixel. So they’ll need to estimate that too.

    Although using all the light is an advantage, subtracting light from some pixels and adding it to others don’t seem that smart. Subtracting 1/3 of the light at one point (assuming same amplitude for all colors) mean that the differnce in intensity will be 2/3 between many neighbouring pixels and they’ll loose maybe half a stop of DR (assuming that the efficiency of the splitter is somewhat less than 100%).

    • All individual pixels might have their own “ISO sensitivity” so it isn’t necessary to have a DR loss due to adding/substracting mechanism.

      • spam

        That would mean that each type of pixel have a different capacity. There might be som clever way to achieve that, but it sound like a really complex design.

        • Sven

          Fujifilm have done this by having different sized ccd cells on the same ccd, I would think it could be done on cmos/live mos too, but I am no expert.

    • Panasonic recently showed a technology for minimizing leaks between adjacent pixels.

    • Esa Tuunanen

      Foveon’s efficiency is sucktastic exactly because of that stacked photosite design without no control where different coloured light is captured.
      While there’s no CFA to filter away light signal gotten from photosites is all mess full of “noise” caused by wrong coloured light.

  • By the way please observe that this document was proposed to be published on 14 March 2012, almost 1 year ago! So it’s highly possible Panasonic to have already developed at least a workable prototype in the meantime.

  • small advice

    Maybe the easier way is to encourage Sigma to work out micro43 body.
    Instead of DP-series with fixed lens Sigma is able to produce micro43 body with Foevon sensor. Such decision can increase sales of Pansonic lenses for micro43. There will be more fans of Foevon sensor if it will be availble in popular mirrorless system.

    Panasonic urgently needs better sensor, because it is behind competition now.

    • Just explained how good is this new concept as opposite to the Foveon-styled stacked layout and you still keep asking for Foveon-sensors in M43 cameras? LOL

    • true homer

      Behind who exactly? And how?

      • small advice

        Sony sensors (used in Sony, Nikon, Olympus and Pentax bodies) and Fuji sensor are better. Competition is not inside micro43 but outside – mirrorless cameras (NEX, NX, Fuji, Canon M) and small DSLR (APS). If Panasonic wants to compete with them needs similar IQ. Specially dynamic range is lower form Panasonic sensors.

        Without better sensors Panasonic won’t be able to compete with others.

        Foevon has one superiority over future Panasonic sensors – it exists now. And Sigma is able to make mirrorless body with Foevon within half of the year (if Sigma will want to do it).

        • I bet the new sensor is quite close to mass production. Don’t forget, the date of publishing request is 14 March 2012, so the quoted document is almost 1 year old! It’s a plenty of time for fine-tuning a such good idea.

          • Gee, Nikon’s three-color sensor patent is even older. Must be even more ready for production ;~)

            What Panasonic proposes, like Nikon, probably requires changes to the stepper process in fab. Doing anything beyond lithography is technically challenging, and it doesn’t always work out in production the way you think (you can get low yields, for example).

        • true homer

          How much more dynamic range than the omd/gh3 you need? It already higher than the models you mentioned.

  • MJr

    So is this going to be in the GX2 ?

    • Bob B.

      Maybe that is why there is such a long delay for version 2. Well….that and finding room to cram a VF in there, (wishful thinking?).

  • “We experimentally demonstrate […]”
    “We believe that the MiCS-IS shows promise as a new method for creating high-sensitivity next-generation detectors.”

    Oops. And here I thought the tech is close to production. But it is still very early lab experiments phase. And the fabrication process is pretty different from the conventional sensors.

    Anyway, the tested circle radius of 1.xx micron indicates that they would first start utilizing the MiCS-IS in small sensors (IIRC 1 micron or less is a typical sensil size for small sensors). What is also logical: mobile cameras outsell DSLR and DSLM by several magnitudes.

    • Just look on the date of request for publishing: the document is almost 1 year (!) old! So the tech can be easily close to production now.

      • jim

        ohhh – I love the optimisum… but a year is not that long – I bet most of this level of manufacture is closer to a 5yr cycle.

        What I hope is Pana can make sensors as good as Sony but also include this so gaining near a stop over the Sony! – unfortunatly all I think this tech will do is make panasonic only very slightly better than sony as sont have 1/2-1 stop advantage as it stands!

  • Renato S.

    If this comes to m4/3 cameras, this will be great news for Panasonic since their sensors have some issues in DR and ISO when compared with Sony, for instance. I wished Panasonic had the time to develop this to launch with the GH3, a custom made sensor for their needs would made them able to get even more in video mode, multi-aspect, maybe 120fps, less rolling shutter, I don’t know.

  • safaridon

    Question – How is this new technology differ from the “smart sensors” of 14MP or 16MP that Pany reportedly has already started production this last December for a smaller sensor as reported in other sources? Just Google it and you may find Pany is already starting to use this technology to be released in future models?

  • So…

    Does this mean Olympus we’ll be running back to Panasonic?

    • The Other Chris

      Why would Olympus run back to Panasonic when all they’d get is old, low-grade sensors from them?

    • Anonymous

      Olympus will never use Panasonic sensors again. They have a good sensor provider in Sony now. Panasonic is a crumb feeding bully. Olympus would at best get G3 sensor from Panasonic, even in 2020.

      • true homer

        You silly people, still thinking the omd sensor is sony…

  • Jón
  • Darryl

    I bet this requires a telecentric lens (light rays impinging perpendicular to sensor). So new life for Four Thirds lenses perhaps.

By continuing to use the site, you agree to the use of cookies. more information

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.

What are Cookies?
A cookie is a small file of letters and numbers that is stored in a temporary location on your computer to allow our website to distinguish you from other users of the website. If you don't want to accept cookies, you'll still be able to browse the site and use it for research purposes. Most web browsers have cookies enabled, but at the bottom of this page you can see how to disable cookies. Please note that cookies can't harm your computer. We don't store personally identifiable information in the cookies, but we do use encrypted information gathered from them to help provide you with a good experience when you browse our website and also allow us to improve our site. You can watch a simple video from Google to find more information about cookies.

Cookies used by our Website
The 43rumors website,, uses the following cookies for the collection of website usage statistics and to ensure that we can . These are anonymous and temporary. By using our website, you agree that we may place these types of cookies on your device.
Read how Google uses data when you use our partners' sites or apps:
Google Analytics Cookie Usage on Websites: Addthis cookies:
Disqus cookies:
Vimeo cookies:
Youtube cookies:

Disabling/Enabling Cookies
You have the ability to accept or decline cookies by modifying the settings in your browser. Please note however that by deleting our cookies or disabling future cookies you may not be able to access certain areas or features of our site. For information about how to disable cookies in your browser please visit the About Cookies website.