[Play Raw] Sunset: Pentax PixelShift ISO 12800 De-noise challenge

License: CC BY-NC-SA

Made this to test noise recovery with the K-1. The JPEG is my attempt with RT 5.6 unstable. There is some motion, but it’s fairly usable?

Auto ISO Noise reduction was enabled, so the JPEG Thumbnail generated by the camera is also worth comparing.

I like to shoot in low light settings, so i was wondering how far i can push it with the K-1.

Setup:

  • Pentax K-1
  • Sigma 50mm ƒ/1.4 EX DG HSM
  • Tripod
  • Bridge with car traffic
  • End of the sunset, almost dark

IMGP0099.edit.iso12800.jpg.out.pp3 (11.4 KB)

IMGP0099.DNG (~200MB)

3 Likes

No noise reduction, but used the median of the 4 frames, cropped to 16:9 and downsized.

Hint: Using Smooth transitions value 0 treats the whole image as motion. Then enabing Use median for moving parts gives the median of the 4 frames:
grafik

IMGP0099-2.jpg.out.pp3 (12.2 KB)

I honestly cannot improve @heckflosse’s result, except for enabling some chroma noise reduction. That gives a very nice homogenous grain to already very nice photo!

1 Like

How does this median compare to the “Median Filter” of the Noise Reduction Module? (if they can be compared)

This median is the per pixel median of the 4 demosaiced frames of the pixelshift raw. The median filter in the noise reduction module is a median on small tiles (3x3 or 5x5 … areas) of the image. Totally different things.

Edit: for example the green light on the buoy (and the reflection on water it produced) in your image is eliminated by using the median of the 4 frames, because this (blinking or rotating) light was only in first frame.

My try with RT 5.6 + DxO

RT

0099.RT.1.jpg.out.pp3 (11.9 KB)

DxO

I know we’re asked to post smaller resolution images, but I can’t really see what you’ve done. I think, in this case, a full res jpg is acceptable. Can you (only the DxO one)?

Following the advice of @heckflosse about demosaicing, I’ve added I pretty good amount of noise reduction:

RT 5.6
IMGP0099-1.png.out.pp3 (11.4 KB)

I wouldn’t go for much more noise reduction, though, as it adds atmosphere to this shot (to my taste).

EDIT to correct the reference to @heckflosse

2 Likes

Where did I advise that? Anyway, your result looks very good :+1: Very clean regarding noise, though still having details.

:cold_sweat: Sorry, I didn’t explain it right: following your advice about the demosaicing…

Previous post edited to correct the mistake.

1 Like

This is a nice raw to shed some light on pixelshift and why it fails here when using plain pixelshift combine.

Let’s have a look at the 4 frames. There’s not much motion, but the light is changing.
Top left (1), bottom left (2), top right (3), bottom right (4), all demosaiced using amaze

As you can see, there’s

  1. the buoy in first frame, but not in the other frames
  2. the light on the right side of the building changes greatly

While a plain pixelshift combination greatly reduces the noise, it also gives some artifacts on the building as you can see here (left is amaze demosaic from first frame, right is pixelshift combined 4 frames without motion correction)

Even the motion correction can’t do much on this dark high ISO shot. Maybe there is some room for improvements, but it was designed for low ISO, where details are not (or less) affected by noise.

Fortunately, there is the median option, which also greatly reduces the noise. It has some donwsides (you may not want to use it if you have large movements in the scene, as moving foliage), but in this case, where we have changing light, and the movements are mostly on the water it works just fine.

3 Likes

Sure , here is the full res jpeg.

In fact, I didn’t do so much as the screenshot shows ( that being said , there is a lot of thing going on under the hood with this prog ) . Did not touch the White balance. Only minor adjustment to Exposure Compensation - Smart Lightening - and Shadow lifting. Did use PRIME Noise reduction.

The noise recovery of the sky gradients is really impressive in @XavAL’s edit.

Thanks @heckflosse for the detailed explanation.

I shot a small series of the scene, at various ISO and aperture settings. This was the most extreme case. I will share some more photos eventually.

You’re welcome!

It would be interesting to see the same 1:1 crop of the right side of the building from the DxO processing by @DxO-user and the RT processing by @XavAL side by side.

Do you mean the same crops as your screen captures? If so, I think a straight away crop of my edit won’t be so impressive, because I developed it to be downscaled (one of the wonders of downscaling is that it hides lots of artifacts :slight_smile: ). But anyway, here it is (without the post-resize sharpening):

IMGP0099-2

If I’m allowed to, I would prefer to compare a quick edit meant to be seen at 100% size (that is, original size):

IMGP0099-3

(what has changed is the global intensity of the wavelets, the chroma intensity of the residual image in wavelets, and added a bit of RL Deconvolution sharpening + Microcontrast)

IMGP0099.DNG.pp3 (12.0 KB)

1:1 crop

With the consent of @DxO-user, I have used a crop of his image to make this comparison image between RT and DxO processing:

The image on the right is the tweaked image to be seen at 100%, as explained in my previous post.