I’ve been getting familiar with my new camera and ran into something that looks like a combination of discolouring and a moiré pattern in one of my images. After some digging and trying I do believe that this is a false colour demosaicing issue, one close to what is described on RawPedia
Switching to AMaZE or AMaZE+Bilinear partially resolves this, even with the false colour suppression steps set to 5 there’s still a remnant of the green visible:
Question: Is this indeed a False Colour Demosaicing issue or am I overlooking something else? Also: can this be refined even more?
RawPedia mentions that, in general, this is seen more with a lack of an anti-aliasing filter. This camera does have one of those…
There’s just one image that shows this in a set of 6, which where shot in a short (25 sec.) time frame. Same bird, distance. Slightly different settings and poses. The others are fine.
PS: I also checked to see if this is an issue in darktable and it is. An obvious moiré pattern is visible. So there’s definitely something in this shot that demosaicing algorithms do not like…
Okay, I was looking for something more specific. Which method suffers the most from this issue: AMaZE or RCD? Your samples show RCD+bilinear and AMaZE+bilinear with colour suppression. It would be better to judge AMaZE and RCD by their merits.
I’ve checked all the “normal” methods before posting (AMaZE, DCB and RCD and their +XYZ versions).
About those I showed: RCD+Bilinear is the default in my auto applied base profile and seems to be showing this issue the most. AMaZE is specific for high frequency detail so I switched to that one, which made it somewhat better. Increasing the False colour suppression slider removed most (“all”) of it. All demosaicing methods do show this and the differences aren’t all that much to be honest.
Okay, so RCD+Bilinear is worse than RCD alone. I suppose dual demosaicing could be tweaked a little more… Do you know which version of RCD this is? I recall there being improvements to the original but I don’t remember whether version 2 was merged.
This was originally done with a somewhat older appimage (dev_5.8-2948-g6189a0e_20210524), just tried the latest build (5.8-3033-gd22ba256d): Same results.
I don’t have a clue which RCD version this is, but this isn’t RCD specific. As mentioned: All show this. Some a bit more or less then another.
EDIT: Remember; This also shows up in darktable. Also using its other methods, some show it more and some less. Doesn’t look the same though.
Just a thought… In my experience birds feathers sometimes show such moiré which might lead to such effects depending on lighting and on camera/sensor/filter. I made a simple test with irfanview (which shows, as far as I now, the embedded jpg). The moiré can also be seen very slightly without beeing demosaiced.
@mozzihh : You might be right, but this is the first time I’ve seen this myself.
This could be one of those one-in-a-million situations where all factors are just right (or wrong if you prefer…). This being a new camera I just want to be sure that this is one of those exceptions you just need to accept.
I’ve extracted the full jpeg using exiftool and do see what you see. The interesting thing is that if I look at the NEF at 100% in XnView I do see a green sheen too. The companion images of this bird do not show this at all.
It seems it’s a known «feature» of the Z6 that it can show moiré in some pictures.
You can get this:
with some strong processing. But to me, the main problem is not the bunch of false colors generated by the demosaicing algorithm, but the moiré present in the raw data.
This is what the green raw channel looks like (the other channels look similar to this one):
Those waves in the feathers are not present in the actual feathers. I bet the moiré is caused by a precise angle of the right details size in the feathers, not covered by the OLPF (anti-aliasing filter). In other words: the details in the feathers have a structure that closely matches the sensor pattern, but are positioned in an angle that the sensor is unable to resolve. This is present in the raw file, and it’s very difficult to deal with by any demosaicing algorithm (but IGV, mostly).
The processing done in the first image is this: Z62_0679.nef.pp3 (14.2 KB)
Most probably there will be color details somewhere else in the image that have been lost, but I was mostly worried by the moiré+false colors.
@ggbutcher : have you ever seen such problem in your Z6?
Makes sense that if it is present after applying any demosaicing method that it would be a hardware issue. However, certain methods tackle moire better.
Regarding RCD, let us ping @LuisSanz. Regarding moire and false colour correction, let us ping @Iain.
I did assume as much, but just wanted to be sure. Although I’ve only put this camera through its paces for 5 days and a few hunderd (test) shots I do seem to notice that it is capable of registering (much) more fine detail then my D750. And I do think that this finer detail capability might run into the occasional moiré pattern if the scene circumstances are just right wrong.
This isn’t an after-the-fact applied contrast issue. Doing so made it more visible, but it is there on a RAW with a neutral profile too. Admittedly the scene itself is rather harsh, but this was intentional and one of the many test I’m doing to see how this camera holds up and what it can('t) do.
LOL I would have expected differently to be honest…
RawTherapee forces you, with some of its modules, to zoom in to 100% to get/see the best results. I kinda like that.
Thanks everybody for your replies and insights!
EDIT: I just had a look at the 20 or so test images I shot this morning from the same bird, in the same location and same harsh light: No problems whatsoever.
Just to illustrate it’s a camera problem of sorts, here is the image open in the develop module of Lightroom at 300% mag - last weeks updated version with camera support!
@XavAL : I had some time to give your sidecar the time it deserves. You do end up with some loss of sharpness and colour detail but it is a nice way to save an image that has this issue if it really counts. I would not have though to try IGV to be honest.
As also shown by @Andy_Astbury1 (Thanks!), this is basically a camera problem. I’m just hoping it isn’t one that shows up too often…
Without a doubt it’s still there, it’s that I haven’t noticed it in the reduced-sized renditions I’ve been producing.
In rawproc, I usually have the export tools (resize, sharpen) at the end of the tool chain so I’m looking at what I’ll get in the JPEG for web viewing as I scooch the tone curve or other prior tools in the chain. I imagine I’ll be seeing more of such if ever I start making renditions for printing, where the output rendition will probably retain the full dimensions of the original capture.
LMMSE and IGV are the winners to resolve the one pixel pattern showed by the barbs.
AMAZE and RCD are completely at loss
To mitigate the problem, I use the defringe tool in RT/ART. It cannot eliminate the moiré that is present in the photo but it can tame the green/red color artifact.
FWIW, the above LMMSE/IGV result isn’t that far off from Adobe’s new AI/ML based demosaicing and upscaling. The vertical stripes are there too and the fine detail is resolved as well. However, the areas of fine detail are ‘overcorrected’ so to speak, they seem like patches of extremely high frequency detail pasted on a regular picture. In that regard, Raw Therapee’s result is actually the more natural-looking one here.
