False Color and Bleeding Example

[JackH]

I would like to enter into the record and share in the most open way possible the following raw NEF, which puts under severe strain false color suppression during raw conversion.

Here is an animated GIF showing how Nikon’s Capture NXD, Photoshop Elements 2020 and Capture 1 Express for Nikon deal with it, best viewed at 100%. Files were rendered with each converter’s Nikon Standard profile, white balanced on the white sign, cropped and exported as 16-bit TIFFs. The TIFF crops were then imported into PS CS5 layers, aligned and resized 400% nearest neighbor for comparison. Note the obvious false color and bleeding in the fence and the red in the sign.

AMAZE and RCD via ART with 5 False Color Suppression Steps were middle of the pack, though I was not able to include them in the comparison because I had difficulty importing them into PS (post on that elsewhere on this board).

Jack

[heckflosse (Ingo Weyrich)]

@JackH I tried to download the NEF, but failed…

[paolod (Paolo)]

The IGV demosaic in RawTherapee works pretty well to get rid of the false color. This is with 5 color suppression steps and capture sharpening turned on with default settings.

[Sunhillow (Hansgeorg)]

IMHO default Amaze, here using 4 false colour suppression steps, is pretty perfect.
I almost never look at more than 100%. 400% does not make sense to me

200%

100%

@heckflosse download worked pretty good

[Thomas_Do (Thomas)]

Darktable 3.4

DSC_2161.NEF.xmp (8.0 KB)

[paolod (Paolo)]

It seems like the false color is much worse in the leftmost side of the fence which is more horizontal vs. the rightmost side which is more diagonal.

[JackH]

@paolod yes, IGV seems to do a good job overall

@Sunhillow, so does AMAZE in RT, better than C1 and PS. I did try AMAZE via ART (all defaults) but it produces more color bleeding than what I see above via RT, It must be some of the other default processing that @agriggio performs under the hood.

@heckflosse, were you able to download the file? Permissions look ok on my side but I am no expert.

[JackH]

[agriggio (Alberto)]

any difference between ART and RT demosaicing is due to white balancing. ART uses the currently selected WB to perform demosaic, whereas RT uses either “as shot” or an internal auto WB (depending on a user parameter IIRC). It seems that this works better here (but this is not a general rule).

[agriggio (Alberto)]

btw this is what I get with ART Amaze and as shot WB (sharpening turned off, I don’t think you want to include that in the equation…)

[gaaned92]

I have some difficulty to see a difference. So going to a test!
RT, ART, who is who? (full scale cropped photos)

photo x:

photo y:

[Sunhillow (Hansgeorg)]

@JackH I think what you see as colour bleeding are just the fine warp threads whose thickness is just below the resolution of the photo.

[JackH]

Good suggestion Sunhillow, though I am not sure because it looks like other temporary fences that I have seen around there and those are made of plastic.

Here are ART RCD, RT AMAZE and PSE2020: I think RT AMAZE has it mostly right, though with the most aliasing, see also around the red sign. So I would think that either during demosaicing or other filtering (e.g. chroma noise reduction) the other two smear things more.

[JackH]

@gaaned92

I have some difficulty to see a difference. So going to a test!
RT, ART, who is who? (full scale cropped photos)

photo x:

photo y:

y is more smeared so that would be ART?

[JackH]

@agriggio

btw this is what I get with ART Amaze and as shot WB (sharpening turned off, I don’t think you want to include that in the equation…)

Yeah, that’s what I mean, there is visibly more color ‘bleeding’ (for lack of a better word) than in RT’s version, though jpeg compression could be contributing to it. To recap, this is what I do to compare the renders, admittedly pushing things to the limit, but that’s what you guys are known for right?

Open raw image at default settings, white balance off the white sign, select crop and export 16-bit TIFF to CS5. In CS5 resize 400% using the Nearest Neighbor algorithm, save as PNG.

[JackH]

AMAZE: ART to the left, RT to the right. Admittedly small differences when viewing at 100%: in ART the snow through the fence appears more contaminated by green and the white around the hand by red.

[Sunhillow (Hansgeorg)]

@JackH

Good suggestion Sunhillow, though I am not sure because it looks like other temporary fences that I have seen around there and those are made of plastic.

At least in Germany they are mostly made of soft plastic coated fiber. I don’t know if cotton of plastic fiber. And between the vertical “bars” there are warp fibers which give can the snow behind the fence a green tint.

Here I found something that might look similar to your fence:

[agriggio (Alberto)]

There’s certainly a visible difference. Can you share the settings you are using (arp and pp3)? Thanks!

[JackH]

Here they are:

DSC_2161.NEF.arp

DSC_2161.NEF.pp3

FYI, with respect to the other issue on opening TIFFs in PS, RT shows 600 dpi in the Crop tool.

[agriggio (Alberto)]

Thanks. (BTW, you can just attach the arp/pp3 here if it’s more convenient).
You have different settings, that’s why you are seeing differences:

• In ART:

[Sharpening]
Enabled=true
Contrast=20
Method=rld
Radius=0.5
Amount=200
Threshold=20;80;2000;1200;
OnlyEdges=false
EdgedetectionRadius=1.8999999999999999
EdgeTolerance=1800
HalocontrolEnabled=false
HalocontrolAmount=85
DeconvRadius=0.62839657068252563
DeconvAmount=100
DeconvAutoRadius=true
DeconvCornerBoost=0
DeconvCornerLatitude=25  

• In RT:

[PostDemosaicSharpening]
Enabled=true
Contrast=20
AutoContrast=false
AutoRadius=true
DeconvRadius=0.63
DeconvRadiusOffset=0
DeconvIterCheck=true
DeconvIterations=20

If you want to compare demosaicing, it’s better to turn off sharpening. I get identical results between ART and RT when doing that.