0
\$\begingroup\$

I've recently switched to a new system (Canon R5 + Canon RF 24-70mm f/2.8L) and I'm experiencing some unexpected noise artifacts when I apply lens profile corrections in Lightroom Classic CC (v12.2).

I've noticed that, for images taken at high ISO (6400+), and hence with significantly higher noise levels, the deformations applied by the profile correction produce aliasing artifacts and create visible rounded/ellipsoidal bands.

These artifacts are only slightly noticeable in the unedited RAWs in Lightroom, but if I try to recover shadow details in dark areas they start to pop up really hard. One would think that dealing with noise is the key here, but I still wonder whether this something to be expected/normal?

Here's an example:

(1) Image with profile corrections enabled / noise artifact visible in the bottom part, where shadows have been pulled significantly. Scaled down to 1920px (height) and still visible. No sharpening applied. Image with noise artifacts and profile corrections enabled

(2) Exact same image with profile corrections disabled / Same scaling, no sharpening. No trace of the artifact indeed, but also unnoticeable noise at when viewed at 100%. Image without noise artifacts and profile corrections disabled

\$\endgroup\$

1 Answer 1

2
\$\begingroup\$

It looks like two of the different things lens profile corrections do are interacting to create the pattern.

Peripheral Illumination Correction

Part of what lens correction does is boost the brightness of edges and corners to offset the drop off in brightness away from the center of the lens. The further away from the center of the field, the dimmer the lens will get. This is more severe in lenses with wider angles of view than in lenses with narrower angles of view. Peripheral Illumination Correction proportionally boosts areas based on how far they are from the center of the frame.

In the case of your sample photo, there is virtually no information from the actual scene in the dark areas at the bottom of the frame to be boosted. Pretty much everything is uniformly dark right around the noise floor.

I'm going to go out on a limb here and guess that you might be using a "protective" filter to prevent scratching that expensive RF 24-70mm f/2.8 L? Or maybe it's just due to the lens having 21 elements in 15 groups?

Shooting in a dark environment with a few bright lights and a lot of dark areas in the frame is the last place you'll get away with using a flat filter in front of your lens (or maybe an unfiltered lens with 21 elements in 15 groups - that's a lot of surfaces for potential reflections). Digital sensors are too perfectly flat and will reflect bright light sources back towards the lens. A flat filter at the front, or even a relatively flat element inside the lens, will then bounce the reflection right back at the sensor. Even multi-coated lens elements are not perfectly non-reflective. Shine enough light on them and you'll see the reflection in dark areas of a photo. This type of lens flare is known as ghosting.

enter image description here

One of the unintended consequences of today's insanely sharp lenses is that they are too sharp for some shooting situations when camera and/or subject movement will blur the image a bit anyway, but the razor sharp internal reflections inside the camera will not be affected at all by camera or subject movement, nor even by missed focus.

Lens flare is always present in images. Usually the entire field is bright enough to mask most relatively weak flare. But when you boost brightness enough, either by shooting brighter or raising it in post, flare in the darker areas of the image will come bubbling up out of the shadows.

The following two images were shot in the same night with the same camera and lens (no filters but the 70-200mm lens has 23 elements in 19 groups and the 2X TC had another 7 elements in 5 groups) pointed at the same spot in the sky occupied by the Moon and Jupiter. The second was exposed approximately 12 stops brighter than the other. The first is also cropped tighter than the second, which is also oriented 90 to the right. (More details here)

enter image description here

enter image description here

Peripheral illumination correction is amplifying any flare and reflections, the amplification is greatest in the corners and on the long edges of the frame, and they are most noticeable in the darkest parts of the frame. When the long edges of the frame are also the darkest areas of the scene, that's when you'll notice it the most.

Distortion correction

Another thing profile correction does is adjust for the lens' geometric distortion. This requires the results from each line in the sensor to be warped into a curve. But the output file can't contain curved lines of pixels, it must describe values for each pixel in an output image in straight lines and columns of pixels. This requires every pixel from at least midframe to the edges and corners to be completely interpolated, as the position of each individual pixel may need to be shifted by dozens of rows/columns. But no two adjoining pixels need the exact same amount of movement.

enter image description here
Imaged sourced from this article and presented as fair use

Notice how each dot is moved by slightly different amounts in slightly different directions? Notice how black dots that were not in a regular grid pattern now are? That also means that the regular grid pattern in the pre-corrected image are no longer in a regular straight-line grid pattern.

Here is a typical interference pattern that can be caused by lens distortion correction:

enter image description here
Imaged sourced from this article and presented as fair use

Look at the right and left edges that would be top and bottom in portrait orientation. Have you seen that type of pattern before?

So now you're putting a screen door with straight wires in front of a screen with curved wires. The pattern you see is the interaction between the distorted shape of one grid (input grid warped to correct for distortion) over the square shape of the second grid (output grid).

Possible solutions

The pattern you're seeing looks to me to be based on a reflection of the sensor's filter stack bouncing off the back of a flat filter, or maybe a flatter lens element, and interacting with the actual grid of the sensor and the CFA on top of it. This reflection is being amplified by peripheral illumination correction. Distortion correction is then creating an interference pattern due to the way the straight grid of pixels that collected the image is stretched and then remapped to another straight grid for output.

If you have the ability use lens correction settings to enable or disable each individual element, try turning off distortion correction while still allowing CA, color blur, etc. corrections to be applied. Or try turning off only peripheral illumination correction and see if that allows the pattern to become so dim as to not be noticeable.

Correction profiles are written under assumptions about how most images will look, with typical scenes being more uniformly bright across the entire frame.

\$\endgroup\$
6
  • \$\begingroup\$ Oh! Thanks for such detailed explanation, now I get to understand a few more things. I still don't quite get why these ellipsoidal marks appear, however, as I do not use any filter, the rear element is not flat, and--most importantly--I only see this artifact if I enable profile corrections. Am I correctly understanding that the effects you described would be visible regardless of profile corrections? I'm beginning to feel that this is just the result of noise being abnormally enhanced as a result of a poor pixel stretching algorithm in the profile's distortion correction. Could I be right? \$\endgroup\$ Mar 10, 2023 at 17:52
  • \$\begingroup\$ Instead of enabling profile corrections, boost the entire image by 2-3 stops. Of course the brighter parts of the image will blow out, but I'm willing to bet that you'll see the reflection begin to emerge out of the darker bottom of the photo. What the profile correction is (partially) doing is amplifying the edges/corners more brightly than the middle part of the frame. It's also correcting for distortion which forces every pixel to be remapped, which means lines that were once straight (such as the grid of your sensor) are now curved and must be translated to straight output lines. \$\endgroup\$
    – Michael C
    Mar 11, 2023 at 1:20
  • \$\begingroup\$ It's probably this remapping that is introducing most of the moire effect. I'll edit that into the answer. \$\endgroup\$
    – Michael C
    Mar 11, 2023 at 1:20
  • \$\begingroup\$ @CarlesAraguz It doesn't have to be the rear element of the lens. Light can pass through glass in either direction and keep going forward until in bounces straight back off of a flat or near flat lens element. \$\endgroup\$
    – Michael C
    Mar 11, 2023 at 2:16
  • \$\begingroup\$ @CarlesAraguz It's not so much noise that is being enhanced by the pixel stretching, it's lens flare that's being enhanced by the pixel stretching. If the area at the bottom was not so dark in the scene, the light from that area would completely drown out the pattern. It's not necessarily a poorly written algorithm, it's just that pixels must be stretched to do geometric distortion correction. \$\endgroup\$
    – Michael C
    Mar 11, 2023 at 2:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.