Pages: [1]   Go Down

Author Topic: Darktable vs. RawTherapee - Moire Reduction Test  (Read 10927 times)

biker

  • Full Member
  • ***
  • Offline Offline
  • Posts: 198
  • powered by a pair of quadriceps
Darktable vs. RawTherapee - Moire Reduction Test
« on: February 20, 2016, 05:46:09 pm »

Here is a moire test sample image photographed with a camera without an AA filter (Nikon Coolpix A).
The pictures are crops from the original image (16 MPix APS-C sensor).

part1.jpg (from top to bottom)
- original camera JPEG (fine)
- darktable Amaze
- darktable PPG
- darktable VNG4

part2.jpg (from top to bottom)
- RawTherapee AHD
- RawTherapee Amaze
- RawTherapee DCB
- RawTherapee EAHD

part3.jpg (from top to bottom)
- RawTherapee Fast
- RawTherapee HPHD
- RawTherapee IGV
- RawTherapee LMMSE
- RawTherapee VNG4
Logged
Don't wait until the flock moves.
Be first to comment. (Unless you haven't got anything to say.)

sankos

  • Newbie
  • *
  • Offline Offline
  • Posts: 36
  • Why matters
Re: Darktable vs. RawTherapee - Moire Reduction Test
« Reply #1 on: February 23, 2016, 02:42:12 pm »

So what are your conclusions?

In my raw converter tests for the Coolpix A I found that the Capture NX-D is quite good at reducing moire (possibly slightly better than RT with amaze). Another good converter for that is DxO Pro, but you need to watch out with this one because its Moire and CA removal defaults are colour-killers (and it is actually the CA filter that takes care of the moire, from what I've seen). Capture One is another good moire remover. LR, my go-to raw converter for the Coolpix A, is actually quite bad in this respect -- you need to take the brush and apply corrections manually. The most atrocious raw converter for moire is Silkypix DSP7 -- I've no idea how they did it, because the NX-D is just a rebranded and simplified SP, and yet SP7's moire correction is almost non-existent.
Logged

biker

  • Full Member
  • ***
  • Offline Offline
  • Posts: 198
  • powered by a pair of quadriceps
Re: Darktable vs. RawTherapee - Moire Reduction Test
« Reply #2 on: February 23, 2016, 04:24:39 pm »

I've added the original RAW to OneDrive. Just right click the icon a download the .NEF file.
I'd be interested to see results of those commercial RAW development programs, if possible. Without any manual retouching, of course!

Please ignore the mess around the sample image. :-[

To crop developed images, you can use Image Magic like this:
Quote
convert -crop 1164x834+1870+1275 DSC_0043.tif cropped1.tif
To make a vertical set from cropped images:
Quote
convert -append cropped1.tif cropped2.tif cropped3.tif -quality 95 target1.jpg
Or, you can do it manually...

So what are my conclusions?
Based on my original samples, nothing really helps.
Camera JPEG seems to be the least coloured one, but at the cost of sharpness.
Darktable Amaze is a bit sharper with a bit more moire.
RT Amaze would probably be the third best - even sharper than darktable with a visibly stronger moire.
Logged
Don't wait until the flock moves.
Be first to comment. (Unless you haven't got anything to say.)

sankos

  • Newbie
  • *
  • Offline Offline
  • Posts: 36
  • Why matters
AMAZE is best for Coolpix A
« Reply #3 on: February 24, 2016, 09:23:24 am »

So, I've downloaded your raw file and converted it in 7 converters on the Windows 8.1. Since the defaults are vastly different for each of them I've decided to apply a different methodology -- getting the most out of each of them with respect to detail and optical corrections (subjective, I know). I cropped inside each converter and developed the file to 8-bit, uncompressed, sRGB tiff (no output sharpening added). Then I opened them in PSE and saved as quality 10 jpegs (so as to avoid jpeg compression discrepancies between the converters). Here are the conversions:
https://www.dropbox.com/sh/h1qp90hvrf37ek4/AABYcdKoh4neJe5wsy5EJGNBa?dl=0

My notes (list sorted by quality):
  • RT (RawTherapee 4.2.699) is the best. My settings: crop; Demosaicing: amaze with 1 false colour suppression step; CA auto-correction on; custom flat-field file for f/4 aperture; Adobe Standard dcp colour profile with TC and Look table on; Sharpening: RL Deconvolution at defaults except for the radius (0.65); Defringe: custom equlizer setting for magenta and green (that helped a lot!). Note: like LR, the Camera WB is interpreted with some magenta bias (as compared with the more neutral NX and CO WB).
  • PN (PhotoNinja 1.2.7). It looks like it also uses something like the AMAZE demosaicing (it's very similar to your Darktable Amaze version. Mind you: this is a screenshot, so don't evaluate the sharpness of the file. Crucial setting: Demosaicing strength set to Moderate. CA correction was left disabled because its results had a negative impact on moire. Sharpening and NR at defaults.
  • NX (the free Nikon Capture NX-D 1.3.0). Not surprisingly, it looks very similar to your OOC jpeg. There are some mazing artefacts. Settings: it's crucial to set the NR method to Better Quality 2013 and zero the sliders for low ISO; the Moire reduction was set to low but it doesn't do anything so you can ignore this; Lateral CA on; USM RGB 10/3/0; Picture Control: Recorded value (i.e. Sharpness 3).
  • CO8 (Capture One Pro for Sony 8.3.4). I have the Sony version so I can't do anything with the Coolpix A file except to convert it to tif. Otherwise I'd zero the NR as it helps sharpness a lot and would run the CA analysis and used the moire slider (possibly brushing in the effect). I'd also use the LCC tool to get rid of the lens colour cast (i.e. magenta vignetting). So, potentially the result could be much, much better with this converter. Interestingly enough, there appeared the red dot artefacts like in your Darktable PPG conversion, which could imply that CO is using a similar demosaicing algorithm.
  • LR5 (Lightroom 5.7). There are two versions: one with global controls only, and the other with the moire removal applied locally with a brush. There are some mazing artefacts present in both versions. CA reduction was on. Sharpening was set to deconvolution: Amount 25, Radius 0.8, Detail 100. Output sharpening wasn't applied, but it would've helped with the overall sharpness, making it even better than RT and CO. The demosaicing is fairly dirty-looking, but it's visible in 200% magnification and doesn't really matter for everyday usage.
  • DOP9 (DxO Optics Pro 9.5.2). Gets rid of colour moire quite well but leaves a lot of mess and artefacts (both mazing and speckling; the dark and light speckles pretend detail and fool the eye to think the photo is sharper than it really is). It's crucial NOT to use the Moire slider at Auto (here I didn't apply it at all, since it doesn't help much but it destroys quite large areas of colour in real life photos). What helped with the colour moire was the CA removal, but set its size to 2 (using the Auto setting is detrimental to colours).
  • SP7 (Silkypix Developer Studio Pro 7.0.2.1). 7. It's supposed to have the same engine which powers the Nikon NX-D but the results are absolute technicolour. Moire is clearly a problem with this converter when developing real life photos.

My conclusion: AMAZE is the best demosaicing algorithm for low ISO photos from the Coolpix A.
« Last Edit: February 24, 2016, 09:28:28 am by sankos »
Logged

biker

  • Full Member
  • ***
  • Offline Offline
  • Posts: 198
  • powered by a pair of quadriceps
Re: Darktable vs. RawTherapee - Moire Reduction Test
« Reply #4 on: February 24, 2016, 02:40:12 pm »

Thanks for your research! It's difficult to judge but Amaze indeed seems to be the best moire killer and fine details keeper at the same time.
As for SilkyPix - I tried it (the FujiFilm specialised version!) for developing my X-Trans sensor RAWs and results aren't very good also in this case. While colours are almost the same like in camera JPEGs, the demosaicing introduces ugly artifacts on surfaces that are perfectly clean in camera JPEGs.

Back to Nikon Coolpix A. I'm adding a few more results, created by UFRaw.
-UFRaw AHD
-UFRaw Bilinear
-UFRaw PPG
-UFRaw VNG
-UFRaw VNG 4 colours inerpolation
All of them (compared to Amaze) are rather pathetic.
Logged
Don't wait until the flock moves.
Be first to comment. (Unless you haven't got anything to say.)
Pages: [1]   Go Up