Fujifilm Varies Pixel Size in Proposed Image Sensor Design


One of the things you kind of take for granted with a digital camera sensor is that every pixel is the same size, but Fujifilm may turn that idea on its head with a new RGBW sensor design that uses bigger pixels for white and green frequencies.

The camera maker describes the design in a patent application filed with the Japanese government in June 2012 and published last month.

The application is a little hard to read via Google translation, but the basic idea seems to be that since the human eye is more sensitive to green and white frequencies, you emphasize those in the sensor pattern. The end result should be less noise and better color reproduction. Or for the tech-inclined, you swap some extra chrominance noise for less luminance noise.


This wouldn’t be the first time Fuji has shaken up the sensor world. The company’s much-loved X-Trans CMOS sensor design (above) uses an irregular RGB pattern to eliminate moire without the need for an anti-aliasing filter.

Of course, this is just a patent, so there’s no word on if or when this design might show up in a camera, but you have to give Fuji points for originality.

(via Egami via Image Sensors World)

Image credits: From Fujifilm B&W patent application, laboriously hand-colored by Your News Ticker and xtranscolourfilter by LiamUK

  • Dr Andy

    There’s no such thing as “white frequencies.” Do I really get to be the first one to say it?

  • Gabor

    Just a small correction: it is quite inaccurate to say “white or green frequencies”. White light contains *all* wavelengths of the visible spectrum with approximately equal intensities. The “green” (L) receptors in the eye are also sensitive to a whole range of wavelengths, which has significant overlap with the “red” (M) receptor’s range. (The main reason why the human eye is more sensitive to green/yellow: there are two kinds of receptors which will detect it.)

    I suspect this might be an attempt to increase low-light performance. If you remove the colour filters from the sensor, you get about a stop of ISO performance (see monochrome cameras). The white pixels will let all light through (perhaps they won’t have a filter at all), so in total more photons will reach the sensor. There’s a physical limit to how much you can improve low-light sensitivity for a sensor of a given size (or reduce noise): the photon noise (shot noise). So it was expected that someone (some company) would try to let more photons through in a sensor design sooner or later.

    Theoretically the Foveon sensor also does something like this: it doesn’t have colour filters and theoretically it should be able to detect all photons (it doesn’t throw some of them away like Bayer filters do.) It seems that for other reasons its performance is still limited today compared to Bayer filter sensors, but I’m eager to see where that technology will develop.

  • Amadeusz Leonardo Juskowiak

    I guess they mean non filtered pixel – pixel without filter can register moreless any wavelength (UV+visible+IR).

  • D.G. Brown

    This actually reminds me a lot of some of the new mobile displays. I have a Galaxy Note 3 which has “diamond pixels” where the different pixels have different sizes, the general layout being like a bayer filter rotated 45 degrees. After I read up on that, I was actually wondering if that same logic could be applied on the sensor side (especially since there’s already so much interpolating being done by the processor).

  • Kevin Purcell

    This design will most likely appear in a small (cellphone) sensor.

    That said 2×3 arrays and non-rectangular pixels will make demoasiacing a bit more interesting but with 14Mpx cellphone sensors the problems in Xtrans won’t be a big an issue. Fuji I assume will provide the code to interested parties.

    It might also allow PDAF focusing using the Canon method of two subpixels under a single microlens/color filter for the white and green pixels (they’re double sized sensels).

    Note that the RGBW CFA layout has the same “rotate the CFA by 90 degrees” each time you move around the bigger block. Just like Xtrans (keep an eye on the blue sensels) so I presume this is from the same designer or design team.

  • aim54x

    A nod back to the SuperCCD with it’s contrast pixels?

  • Dr Andy

    I assume so, too. I just couldn’t let it pass.

  • OtterMatt

    I’m glad we have such smart people in the comment section, actually. I just had a minor mind-blow moment reading up on the theory and guesses you guys had.
    At least to a layperson like me, the design seems to make more sense than the standard Bayer configuration. Sadly, this will probably never make it to a camera that I can afford.

  • tomdavidsonjr

    Interesting. I posted a comment yesterday questioning this article’s statement that: “the human eye is more sensitive to green and white frequencies”, and indicating that it was a shame that an article on a website specializing in the art and science of light would make such a ridiculous claim (because there is NO such thing as a “white” frequency of light), and asking that the author correct this. Instead, my comment was deleted. Things that make you go “hmmm”…

  • herbst

    It will be named the Fuji X-Mondrian

  • kodiak xyza

    hmmm… yeah. I had a suggestion that the phrase “uses an irregular RGB pattern to eliminate moire” is non-sensical, because patterns, by definition, are not irregular.

    more to the point, the pattern is regular, with the tile as shown being bigger than the Bayer, and that within that tile, there is an easy rotation/reflection for the 4 sub-tiles.

    now I notice that the claim that it does not need the anti-aliasing filter may also be imprecise. it can be that the pattern is more robust to the removal of the anti-aliasing filter, but not the same as “to eliminate moire without the need for an anti-aliasing filter.”

    whatever, but yes… hmmmm.