FX9 Green Fringing and White Balance Update From Sony Japan.

I had a long, very in depth and very interesting web meeting with several members of the FX9 engineering team in Japan last week. They have been looking into the green fringing that some FX9 users have been reporting in various user groups. I have to say that this hasn’t been something that I have found to be a problem, but if you keep on reading you’ll find out why that is!

The first thing that became very clear in the discussion was that they take issues like this extremely seriously. The FX9 is their baby and they want end users to be happy with it. I was shown a lot of examples of very carefully executed tests using calibrated light sources, test charts, and various objects with different colours or reflectivity placed in extreme contrast situations.

And yes, in some cases the images would show a green edge around the high contrast edge of an extremely backlit object. Not just with the FX9 but also with the FS7 they were comparing it against.

The cause is Chromatic Aberration in the lens.

In every case the cause of the coloured edge is chromatic aberration. Chromatic aberration is an optical phenomenon cause by the way the glass in a lens will inevitably refract different wavelengths of light (and thus different colours) by different amounts. So when one colour is in precise focus, other colours may be slightly out of focus and this causes blurring of one colour and this typically manifests itself as a blue, green or red fringe – chromatic aberration.

So the root cause of this issue is not the camera, it is the lens and it will vary from lens to lens. But do keep reading – there is more you need to know…….

Other Factors.

However there are other factors at play. One is the way a bayer sensor works. There is a huge resolution difference between the green channel and the red and blue channels. A lot of complex processing is used to compensate for this and that processing is optimised for what you might call “normal” or “typical” scenes. So when presented with a shot that has extreme contrast and a lot of optical chromatic aberration sometimes this processing isn’t going to deliver a fully optimised image and it may reveal the differences in resolution between the 3 colour channels. This is particularly the case with green CA as the green channel has much greater resolution than the others. 

But the FS7 doesn’t do it….

But this bayer/processing issue will be broadly similar between the FX9 and FS7, so why are more FX9 owners reporting a problem than FS7 owners?

The answer it seems is white balance!

All the  reports of this issue have occured when the camera has been set to a daylight white balance, most of the time it has been when the camera has been set to it’s default 5500K daylight preset.

When comparing the FX9 with the FS7 with both set to 5500K the FX9 has a small but noticeable bias towards green. I and many others have been aware of this from the launch of the camera, especially when you use the camera in the CineEI mode and use the standard s709 LUT from Sony. The images are distinctly more green than any other Sony camera and a fair bit more green than the FS7, FS5 and Alphas.

It’s not that white objects won’t be white, it’s just that overall the image looks a bit green. This is easily dialed out by adding a +20 offset to the cameras tint control and that is how I have been using my camera pretty much from day one.

When you think about it – or when you analyse the situation carefully as Sony have, if the camera has more green, what will happen if you have green CA? Well it’s going to be much more pronounced.  And that is what we are seeing. The way the colours in the FX9 are tuned has an unfortunate side effect of making green CA much more obvious.

This explains why I have never really had an issue with green CA. I’m normally running a +20 tint offset which reduces the green in the camera plus I tend to use my own less green versions of the s709 LUT or I use ACES which is also less green than Sony’s own s709 LUT. So overall I have already dialed out the green and unknowingly fixed an issue I didn’t know I had.

So the engineers in Japan are currently recommending adding a +20 tint to the camera when shooting daylight. In particular instead of using a preset of 5500K to use 5000K and +20 tint to achieve the same white balance but a less green image. They are also assessing whether the camera needs to have it’s processing adjusted so that it is less green when using the 5500K preset. At warmer colour temperatures this issue does not seem to ever be a problem, it’s only with daylight.

Push Auto White Balance.

Further to this it’s also worth noting that if you use the cameras Push Auto white balance to set the white balance with a white or grey card then this will normally deliver the optimum white balance for the scene you are shooting and it won’t be biased towards green. And we should remember that Push Auto white balance works correctly in both the CineEI mode and Custom mode. If you use Push Auto WB there is no need to add a +20 Tint and the pictures should look natural and well balanced. Having become used to using preset white balance values when shooting S-Log with Sony cameras for so many years I keep forgetting that you can now do this. 

Another benefit of Push Auto WB is that as it balances the camera correctly it also helps bring the FS5, FS7, F55 etc and FX9 closer together if you are using a mix of different cameras. I was recently shooting with a Venice and FX9 side by side and by using the Push Auto WB on both Venice and the FX9 it became extremely hard to tell one from the other. Previously when using the presets the FX9 always looked a touch green.

Also – if you are using daylight balanced LED or fluorescent lights using Push Auto white balance will help correct out any tendency towards green that the lights may have. So really Push Auto White Balance is a win-win situation and let’s face it a lot of you have been asking for it in CineEI for a long time – so now we have it, let’s use it.

What about Aperture Correction?

In my own experiments I found that in Custom Mode turning off the cameras Aperture correction (http://www.xdcam-user.com/2020/07/reducing-ca-artefacts-in-the-sony-fx9/) can help reduce the visibility of the CA. As CA is an optical edge effect, anything that enhances edges will also enhance the CA and turning off aperture correction prevents the CA from being boosted by the aperture correction. Overall I personally prefer the images you get from the FX9 with aperture correction off anyway, they are a little bit less sharp and more rounded. 

And slightly off topic, what about image quality at FF and S35 2K scan? This was discussed as well, the engineers want people to be happy with the camera. The issue is that this is a camera with a 6K sensor and an optical system designed around that 6K sensor. So if you then reduce the resolution of the sensor readout the optics are no longer optimised (mainly the Optical Low Pass Filter). As a result there is an inevitable increase in aliasing and moire. Unfortunately this is just the way the physics works and there is not much that can be done about it. But the engineers know that the FX9 will be compared with the FS7 – which also has similar issues in it’s S35 2K scan modes. The goal of the engineers is that the FX9 should not be worse than the FS7 and there are some tweaks in the pipelines to the image processing that will bring some improvements to the FX9 when using 2K scan. But let’s be realistic, this will always be a camera with a 6K sensor and a 6K OLPF, so the Full Frame and Super 35mm 2K scan modes will never match the quality of the 6K and 4K scan modes, it just isn’t possible so don’t expect miracles. These tweaks may take a bit of time to be finalised, so I’m not sure when we will see them. I think what we should see however is the 2K scan from an FX9 being indistinguishable from the 2K scan of an FS7.

22 thoughts on “FX9 Green Fringing and White Balance Update From Sony Japan.”

  1. Thanks for the update Alister. I do wish Sony would issue regular updates letting us know what they’re looking in to, I think it may help reduce some of the online hand wringing and anger.

  2. Does the +20 tint (anti Green) apply for custom S Cinetone too .. or this is just an Slog/ Cine EI mode thing ..

    Thanks

  3. Thanks Alister.
    Very helpful post.
    Just curious, is the Venice also seeing the same green effect through CA like the FX9/FS 7

    1. CA is CA and if the lens produces CA it doesn’t matter what camera you use it will still be present. With Venice how obvious it is will depend on how you grade the material.

  4. In preset S-Cinetone you don’t seem to be able to adjust tint? Only by dialing in the colour temp and tint in A and B?

    Cheers!

  5. I have always white balanced the FX9 and STILL get horrible fringing and cyan pixel creep along edges in my images. After getting back from a 2 week shoot in Arizona, I am having to toss about 5% of my shots (Cine Ei/SLOG3 to PQ/ST2084). I am using SONY GM lenses and an S4 Cooke. I am having a little trouble swallowing the CA theory. I will give the 5000K and +20 a try. For now, as a former FS7 owner – I am not a happy FX9 customer.

    1. CA is a fact of life. No lens can be CA free because lenses rely on refraction and refraction is frequency dependant and thus colours will be split as they pass through the glass. As the quality and resolution of cameras increases CA will become more and more challenging. The FX9’s oversampling is working against it as you have 30% higher resolution so any lens imperfections will be much more visible.

      I’m surprised you not seeing this while shooting and have to discard footage in post? Are you using the Loupe or just relying on the LCD on it’s own and perhaps as a result unable to actually see fully what you are capturing? With proper monitoring there should be no reason why you won’t see this while shooting so you can adjust the aperture or re-frame to avoid the extreme contrast scenarios and aperture ranges where CA will be at it’s worst.

      1. Thank you Alister.
        I am just wondering why I don’t see it with my A7III and A7SII. I am also curious if the ND assembly (ND filter and/or protective glass) in front of the sensor has any effect in refracting light into CA ? You would know more about that than me. I will do some testing this evening.

  6. The Alpha cameras use a different debayer process to the FX9 and this does have an effect on how visible the CA is, but the CA will still be there. Sony are working to improve the processing to further minimize it but ultimately the cause is the lens and the physics of how glass bends light.

    The ND filter or any other optically flat glass should not introduce CA as they are not bending the light in the same way that a lens does. In a lens when you bend the light by refracting it each wavelength bends by a different amount, but as a flat glass should not be refracting the light it should not introduce CA.

    This is why we don’t normally see CA when looking through a window. Zoom lenses are almost always the worst lenses as they have a more complex design with a lot more elements doing a lot more bending. Larger lens elements can be more prone to CA as they tend to also be thicker and aperture also plays an important role. Further sensor size matters as the larger the image circle the larger the angle of the light at the lenses exit pupil and this magnifies any errors.

    Another factor is that as the dynamic range of a camera improves we are tempted to shoot in ever higher contrast situations as we expect the camera to deal with it, while with a camera with less DR we may not even attempt the same shot. More contrast = more problems.

    Generally lenses use a mix of different types of glass with opposing refractive indexes to try to minimise the amount of CA (normally called apochromatic) as well as modern cameras using electronic compensation such as ALAC.

  7. Alister,
    Now that the new firmware accepts User LUTs will you be making a modified S-709 LUT that alleviates the green problem? (I’ve sometimes found that I liked the old LC709A LUT better with FX9 because it wasn’t green and actually liked S-709 with the FS7 because it was less magenta . That switch also seemed to facilitate color matching those cameras although the gammas were a little different . Now that user LUIT’s are available I’m hoping that match will become easier in the field when in CINE EI . ( You did explain how those 2 will never match perfectly though.)

    1. The green is part of the s709 look. It’s there with Venice too. It is designed to in part mimic many film stocks that do not result in a perfect real world representation but a slightly stylised one. LC709A will be more true to life as it is closer to the vanilla 709 spec of your monitor, so will be more accurate. This is always a trade off, accuracy, which may not always be considered pretty or a more stylised look.

      1. Thanks for that explanation. I understand that if you balance under green fluorescent light, your grey card may appear correctly, but skin tone would read green, because of the light interacting with reflectance and bayer spectral sensitivity. However, in this case, I’m not asking the camera to be “correct” to any real world color other than the grey card, so I’m still a bit confused as to how the camera arrives at an incorrect value. Maybe I’m not understanding how it calculates WB? As you say, it’s looking at the cumulative values for R, G and B channels and attempting to balance them uniformly, so for a grey card at a fixed exposure (even if nothing else on set) shouldn’t it always arrive at 50-50-50?

        Thank you so much for your patience in educating us, you’re truly a gem of the internet!

      2. Thanks as always Alister,
        Interesting observation. I guess how much green is “stylish” vs “GREEN!” depends to some extent on taste. I may need to pull those rose colored glasses out of the desk again. Wouldn’t hurt for life to look a little prettier these days anyway.

        Lenny

  8. Hi Alister,
    I did my own testing with a vectorscope and signal analyzer with the camera in CineEI, pointed at a light source and exposed at 50%. After running push-auto, the camera still needs further offset. Testing with four data points between 2700 and 6500K, the camera needed between 5 and 11 points of +M tint (average 8.5) and between -60 and -446 (average -211) kelvin to reach a perfectly grey value. It’s unfortunate that those values aren’t consistent for every white balance, instead the auto WB is less accurate (reads more and more warm) at higher white points. 500K is a noticeable shift.

    1. And you used a calibrated full spectrum white light source for your tests or an LED panel with a far from uniform full spectrum output?

      1. I tested the unit in question with a Sekonic C800, it gives a TCLI of 97 and a CRI Ra of 94.6. Should that matter to the auto WB? In this test, I’m not comparing the values the auto WB reads to the real-world values on the meter, I’m comparing the values the camera sets to a manually set value in the white menu that brings the video signal to dead center.

        If the camera is presented with a frame that consists entirely of one color, and we ask it to auto WB, shouldn’t the camera’s goal to be to get the signal to perfect grey, no matter what spectral spikes it’s presented with? Virtually any broad spectrum light source (ie not sodium vapor) will have a point in the white menu settings that will center the vectorscope.

      2. The camera only has R, G, and B filters, it’s not a spectrometer, so the result of auto WB on an all-white frame shouldn’t be affected by spectral spikes, the result should always be grey. I can understand how it would get it wrong if we were dealing with reflectance off colored objects reacting with spectral spikes in the source, but I’m pointing a 200mm lens directly into the panel in this case. Please do correct me if I’m misunderstanding how auto WB works! What’s mysterious to me is how it arrives at a different amount of offset per color temperature—I suppose what’s happening is that the camera reads raw values off the sensor, and then consults a chart that maps X input to Y CCT and tint, and that chart is further off the higher you go. I suppose that makes sense if the default values are 3200K?

        1. Absolutely, spectral spikes will skew the white balance. In camera white balance is purely a gain function so is based on total cumulative amount of light in each channel with the assumption made that the light source is absolutely uniform. It is a very coarse adjustment. The fact that the FX9 now in addition has atint control gives a bit more “resolution”, but you still only have 4 correction points. As a result it cannot correct for the holes and spikes that LED’s produce, so often the result will contain a tint because the gain levels are correct but perhaps there is more green in the Red end of the Green or less green in the blue end of the green and WB cannot address this. The camera can only ever work with a single average value for each of R G and B and if R G and B are not uniform there is a high probability that a tint will remain as the white balance and tint corrections are too coarse to compensate for the lights imperfect peaks and troughs. We used to see this with fluorescent lights all the time where white balancing under fluorescent lights would leave a green tint and the only way to deal with this was with a magenta filter or adapted colour matrix. The closest analogy I can come up with is audio. If you want to correct imperfections in an audio system with just treble, mid and bass adjustments you will get somewhere close, but really you want to break the range down into much smaller chunks with a graphic equaliser if you want to even out or eliminate smaller peaks and valleys.

          The different offsets are what you would expect to see at each colour temperature as the sensor itself has a fixed colour response, so as you move away from the base R G and B gain levels you will see ever greater offsets in R and B.

        2. But “white” light from an LED light isn’t really white. 10 different cameras will all reproduce the light from an LED light 10 different ways, each giving a different white balance depending on the overlap between the colour filters and the LED’s peaks and valleys. White balance is crude, coarse, R, G, B gain adjustment that equalises the R G and B channels assuming everything is absolutely uniform and linear so that there is no chroma component, only luma.

          If one of the lights spikes happens to fall right on the crossover between say green and red there may be a very narrow wavelength band, but high volume of red photons hitting the red pixels and that will skew the WB away from red while the overall average value for red channel could be quite low. Then because of this non linearity even the smallest difference in the gain, exposure or anything else will result in a WB shift, even the very process of adjusting the R G and B gain levels to adjust the WB has the potential to skew the WB one way or another.

          Trying to perform assessments of a cameras white balance function using light that is not full spectrum and absolutely uniform will give you unreliable results.

          1. I see, so a push-WB at one exposure value may not be consistent with a reading at a different value. There’s still parts of this I’m finding hard to wrap my head around, is there a good resource somewhere to learn how the AWB is calculated? You’ve been very generous with your time, and I don’t want to eat it up explaining it if I can read up on my own.

            I’ll repeat the test using a tungsten bulb and daylight.

Leave a Reply to Robin Probyn Cancel reply

Your email address will not be published. Required fields are marked *

*

This site uses Akismet to reduce spam. Learn how your comment data is processed.