Project

General

Profile

Bug #10333

X-Trans images with moiré (in DT but not in LR)

Added by Francisco Cribari about 3 years ago. Updated about 3 years ago.

Status:
New
Priority:
Low
Assignee:
-
Category:
Darkroom
Target version:
Start date:
02/19/2015
Due date:
% Done:

0%

Affected Version:
git stable branch
System:
Ubuntu
bitness:
64-bit
hardware architecture:
amd64/x86

Description

I have been noticing moiré in several Fujifilm RAW files (.RAF, X-Trans). Here is an example:

RAW file: https://dl.dropboxusercontent.com/u/2171814/_FFX7922.RAF

XMP (RAW): https://dl.dropboxusercontent.com/u/2171814/_FFX7922_01.RAF.xmp

JPG (SOOC): https://dl.dropboxusercontent.com/u/2171814/_FFX7922.JPG

XMP (JPG): https://dl.dropboxusercontent.com/u/2171814/_FFX7922.JPG.xmp

There is moiré in the blue shirt in the RAW file, but not in the SOOC JPG. (Take a look, e.g., at the area between his camera and his vest after zooming in.) Even with Markesteijn 3-pass demosaic + five times color smoothing I still see moiré.

I wanted to find out whether this is a Fujifilm-related issue or a Darktable-related issue, so I asked a friend to open the RAW file in Lightroom (version 5.7.64) and export a full resolution JPG. (Since all my computers rum Linux, I do not access to LR.) Here is the resulting JPG:

JPG exported by Lightroom: https://dl.dropboxusercontent.com/u/2171814/_FFX7922b.jpg

I see no moiré.

It seems that one needs to use very aggressive denoise profiled (wavelet + color) and defringe settings to get rid of the moiré, as noted by Marc Cousin in the darktable-users mailing list. Nonetheless, that has side effects. I reproduce Marc's latest message:


from: Marc Cousin <>
to: Francisco Cribari <>,
Darktable-users <>
date: Thu, Feb 19, 2015 at 12:57 AM
subject: Re: X-Trans images with moir? (in DT but not in LR)
mailed-by: gmail.com
signed-by: gmail.com
: Important mainly because of the people in the conversation.

I'm still wondering about a problem I've seen very clearly with this RAW: when denoising with wavelets, there are green artefacts on many edges.

Here is the original image. Only demosaic (markesteijn 3 pass, with five times color smoothing), basecurve

https://drive.google.com/file/d/0B_liIOwDeWUmWnVSMVdpcFp2bU0/view?usp=sharing

There is moiré, and small artefacts on the camera's edges, but it's not that bad.

Add a profiled denoising, wavelet, on color. Be a bit more agressive than the default values.

https://drive.google.com/file/d/0B_liIOwDeWUmeTZLUnk1eDBubHc/view?usp=sharing

Moiré is gone from the shirt. But you get green on lots of places… on the camera for example.

Adding a very aggressive defringe is a way to get rid of this green, but destroys a lot of colour...

https://drive.google.com/file/d/0B_liIOwDeWUmXzRBZG53bHdYaFE/view?usp=sharing

But I feel that the defringe shouldn't be needed here (or at least not for that), and there is something I don't get about the denoising…

Could someone shed some light on this ? :)

The original RAW file was here: https://dl.dropboxusercontent.com/u/2171814/_FFX7922.RAF

Regards

Marc


I am not sure this is a DT bug, but I believe it is something that is worth looking at. P.S. I run DT version 1.6.2+17~g15823c5 on Ubuntu 14.04.

History

#1 Updated by Pascal de Bruijn about 3 years ago

So moire is an expected issue in digital photography, especially with sensors that have the low pass filter removed.

Even though Fuji advertises it's X-Trans to be better Moire resistant, there has never been any verifiable proof to this statement. So there is no real reason to think this is actually the case. Fuji has a long history of over-marketing over-complicated sensor designs that often aren't really that much better than a traditional Bayer to begin with.

The out of camera / lightroom images aren't too relevant, as this shows the result of a full imaging pipeline, they may just be employing a general moire reduction algorithm (whatever it may be).

Other tools implementing Markesteijn demosaic for X-Trans show exactly the same moire pattern, so our Markesteijn implementation doesn't seem to have any Darktable specific bugs. It might be interesting to hear Frank Markesteijn opinion on this matter though.

PS: While not perfect either, you might want to try "denoise (local means)", 1, 100%, 25%, 100%, off

#2 Updated by Marc Cousin about 3 years ago

This issue is more about the second part (my response email to Francisco's question): green artifacts produced by the wavelet denoising. They don't feel normal.

There is no question that moiré is a problem we are all facing, and that Fuji's sensor isn't magic.

By the way, this issue is about the problem I have been talking about on IRC today (marco44).

#3 Updated by Roman Lebedev about 3 years ago

Were there any Fujifilm-specific settings enabled when the taken was taken, like a shot ISO in the EXIF != actual ISO that was used to take the picture?
IIRC Fujifilm only records the intended ISO, not the actual one, so in such case dt uses noiseprofile for wrong iso.

#4 Updated by Francisco Cribari about 3 years ago

No special settings. The ISO used was 400. Shutter speed was 1/90s and aperture was f/5.6. The camera settings that affect the SOCC JPG were: 1) noise reduction = -1, 2) HTone = 0, 3) STone =0, 4) Sharpness = +1, 5) film emulation = STD, 6) DR = AUTO. Additionally, White Balance = AUTO.

Roman Lebedev wrote:

Were there any Fujifilm-specific settings enabled when the taken was taken, like a shot ISO in the EXIF != actual ISO that was used to take the picture?
IIRC Fujifilm only records the intended ISO, not the actual one, so in such case dt uses noiseprofile for wrong iso.

#5 Updated by Marc Cousin about 3 years ago

Exif isn't the real ISO: this was taken with DR Auto. The camera chose DR-200, so it means that the ISO displayed is the effective ISO in the highlights. Here, for instance, it is DR-200 and 400 ISO in exif.

It means the picture is taken with the sensor at 200 ISO, underexposed by one stop. To have a correctly exposed picture, in Darktable, you have to set exposure to +1. In order to get a similar rendering to the camera, you also have to either play with shadows and highlights, or modify a base curve or tone curve, for instance.

Anyway, the effective ISO is 400 in this picture, for the highlights at least. And even with the 200-ISO profile, I get the same artifacts. Less of them, of course, as denoising is less agressive, but I still have them. And of course, the moire is stronger, and other color noise too.

#6 Updated by Francisco Cribari about 3 years ago

I keep having problems with moiré in X-Trans RAW files in Darktable. Here is another example (Fujifilm X100S):

.RAF: https://dl.dropboxusercontent.com/u/2171814/_FFX8484.RAF

.XMP (RAW): https://dl.dropboxusercontent.com/u/2171814/_FFX8484_01.RAF.xmp

JPG SOOC: https://dl.dropboxusercontent.com/u/2171814/_FFX8484.JPG

Marc Cousin wrote:

Exif isn't the real ISO: this was taken with DR Auto. The camera chose DR-200, so it means that the ISO displayed is the effective ISO in the highlights. Here, for instance, it is DR-200 and 400 ISO in exif.

It means the picture is taken with the sensor at 200 ISO, underexposed by one stop. To have a correctly exposed picture, in Darktable, you have to set exposure to +1. In order to get a similar rendering to the camera, you also have to either play with shadows and highlights, or modify a base curve or tone curve, for instance.

Anyway, the effective ISO is 400 in this picture, for the highlights at least. And even with the 200-ISO profile, I get the same artifacts. Less of them, of course, as denoising is less agressive, but I still have them. And of course, the moire is stronger, and other color noise too.

Also available in: Atom PDF