Sony A9: Two Examples of Pattern/Banding Noise at ISO 100 (Pine Creek Skyline, Mt Tom) + Reader Comments
UPDATE 5 June: I’ve added a 2nd example shot in uncompressed raw with the Sony 70-200mm f/2.8 OSS GM. The pattern noise problem in the red channel remains the same. Clearly this is a camera problem. Whether Sony can fix it in firmware is unclear. See also the reader comments at end.
Back in May I reported on Sony A9 pattern noise at ISO 100 (poppy). The “bad camera” theory cannot be strictly ruled out, but I have never found that sort of explanation to be true over the past decade with any camera. And if one accepts that theory, then I’d want to very carefully evaluate at $4500 purchase.
One reader states he is unable to see any banding with his A9 over thousands of exposures (update: this reader has now seen the same noise pattern in his own photos). I accept that statement at face value, but only that, meaning there are many possible reasons including ones not even involving the camera (e.g., a bug unknown to me in ACR and/or image processing, the display, or brain wiring). Others (web sites) have not even seen my analysis, but dismiss it by a popularity contest, as if truth or falsehood of any factual claim rested on how many people believe or disbelieve it (the malady of our times).
I state here (as before) that for most color images, the pattern noise that I show is unlikely to be an issue for “ordinary” usage, since the green and blue channels blend with red in a color image (thus masking noise), plus subject color, type of lighting, and exposure also contribute substantially. The Sony A9 makes a very fine camera for what it is designed for (very high frame rate).
Red channel pattern noise
I shoot outdoors a lot, and that often means a dark blue sky, particularly in the mountains and when taking care to not blow out the highlights (snow, bright granite, etc). Dark blue is weak in red channel exposure, so that a pattern noise problem in the red channel should stand out. And so I had planned to approach it on just that real-world basis, which I have now done—and 2 minutes of shooting proved it out. Of course, a bright sky or one overlaid with clouds will give the red channel much more exposure, which will minimize noise and make the pattern noise of the A9 be very subtle to nearly invisible, a corollary that I have also verified.
One of the things I regularly do is to make black and white conversions, and usually the dominant channel used is the red channel. Hence a clean noise-free red channel is critically important. Making matters worse in practice, such conversions are likely to involved high contrast scenes with... blue sky. Precisely the worst case for the A9 red channel.
Presented at image sizes up to full camera resolution. Two examples are shown, one shot with the Zeiss Batis 135mm f/2.8 APO-Sonnar in compressed raw and the other shot with the Sony 70-200mm f/2.8 OSS GM in uncompressed raw. Using uncompressed raw helps in one way but not with the pattern noise.
The first example shows three black and white conversions, starting with RGB images in ProPhotoRGB, AdobeRGB, sRGB. The second example shows the RGB image as well as the red channel in the ProPhotoRGB, AdobeRGB, sRGB color space—with awful results.
As proof of the incompetence of DxOMark to evaluate noise, see their “one of the best sensors we’ve tested” comments. This blindness to pattern noise has plagued DxOMark for years. GIGO.
The image below is the red channel of an sRGB color image. It is greatly reduced in size, which reduces the appearance of noise. Shot in uncompressed raw, the red channel pattern noise is unusable for any type of landscape work I’d want to do.
The image below is a black and white conversion using a high contrast red filter layer (Photoshop). Even at greatly reduced size the red channel pattern noise is still visible.
Dennis F writes:
I don’t and won’t be using an A9 – doesn’t fit my need – but I have seen this pattern on an a6000 I converted for IR (830nm). This same pattern showed up in some areas depending on the luminosity. It was especially noticeable in clouds. I had to stop using the camera.
DIGLLOYD: makes sense in context of the pattern noise in the red channel. It is a good example of why “same sensor” is a ludicrous conflation with image quality : camera electronics have a big influence on image quality which is why I still name the Nikon D810 the finest 35mm camera on the market—Nikon has superb image quality.
I heard that you recently tested the A9 and found horizontal artifacts in the red channel. This is something that we have been seeing in the A7R II camera in the IR spectrum. I attached a sample of our artifact, let me know if this is similar to what you see. It seems like some of the pixels Sony puts has a different IR/red transmission.
... I would bet it's an oversight of the spectral properties of the AF pixel rows. Or large parts of the row, I think. The first time I ran into this was the Canon M3 that had a hybrid focus sensor, that one had horizontal lines that were clearly visible when you looked at the sensor. They might not use every pixel but for production they probably photomask whole rows at a time. s
DIGLLOYD: yes, the effect is very similar, I would deem it the same defect.
Salim M writes:
Thanks for continuous in depth reporting. You may not know this, but I was a life-long Canon user, until I saw your similar reporting on red-channel of Nikon vs. Canon from many years back. It was the last straw that convinced me to switch from Canon to Nikon which luckily coincided perfectly when D800 came out. For me still the Nikon D800 and D810 are the best cameras ever. If the new D820/50 will indeed have a better AF. I don't think I'll ever need another camera again.
DIGLLOYD: let’s hope Nikon moves things forward, including an EVF option in the hot shoe.
Nathan C writes:
Regarding the banding in the red channel on the new Sony A9, I'd just like to point out the similar issues I've encountered with my original A7, and make you aware of another field of imaging that this flaw effectively disqualifies them from
I shoot film exclusively, and I digitise my film using an A7 mounted on a homemade copy stand. I use a Canon FD 50/3.5 Macro (simply superb) at around f8 on an extension tube, and illuminate the negative from below with a twin flash setup. Most of the time it works just fine, however with certain frames I have found very obvious banding in particular image regions, and if I look for it I can discern similar flaws in most images subjected to a lot of colour or contrast manipulation (ie most colour neg scans, some underexposed B&W negs). C-41 colour negative film is of course very orange, and loads up the red channel, and I would volunteer that very blue or green scenes captured on neg film scan poorly with this camera.
It's an issue I never encountered with my Canon 550D, and it makes me think that RAW files from the Canon have much better "integrity" than the Sony files. I process out the RAW files using the godawful Sony Image Data Converter program, with every form of correction or adjustment or denoising turned off, inverted and with the WB "ballparked", and edit the JPG's from there. You may cringe at editing JPG files, however I'm quite confident that what I've seen is camera induced rather than editing induced. I've tried exporting as TIFF, etc. I'll be switching to a 50MP Canon at the earliest convenience in the hope of getting back uncooked RAW files, and I should mention a magnified live view that comes into sharp focus (and battery life, and sane menus, and a real viewfinder, incase I ever want to use it for real photography as well sometimes).
DIGLLOYD: genetic predisposition of cooked raw files? Or just processing electronics that are not as good as Nikon and Canon? At any rate, the D810 is much preferable to the Canon 5Ds R for image integrity (noise and dynamic range)—I have both. The resolution difference is marginal.