7

I took a picture this weekend with my Canon 450D, and I noticed an area with a strange pattern:

strange pattern

What could possibly be causing this? Does it come from the sensor? Feels to me like it's the physical pixels that appear somehow on the picture (maybe because of the position of the light source).

And more importantly, is there a way to avoid/take advantage of this?

EDIT

This is a RAW picture processed with Lightroom 6.7.

inkista
  • 52,065
  • 10
  • 88
  • 161
gregseth
  • 255
  • 3
  • 9
  • 3
    Was this a jpeg straight out of the camera, 100% crop, or if it was raw, what software was used to process it? At least some of what I see there look very much like demosaicing artefacts. – PeterT Nov 14 '16 at 10:52

1 Answers1

9

These are demosaicing artefacts introduced when converting the image from the bayer array that the sensor records into an image with complete colour information for each pixel. In an effort to reproduce as much high frequency content as they can during this process, the filters involved in some of these algorithms sometimes 'ring' in a way that introduces "maze" artefacts at the nyquist frequency (i.e. 2 pixels). The blotchy quasi-periodic red stripes are probably related to moire introduced in a similar way.

This previous question might have more explanation and background in the answer (I haven't reviewed in detail) What are the pros and cons of different Bayer demosaicing algorithms?

If the image is straight from the camera, then I'd say that's a JPEG output bug in the firmware and the only option I can see would be shooting raw instead (or checking if there's a firmware update). Otherwise, you need to pay careful attention to the raw conversion software you're using and the settings you apply.

At the risk of casting aspersions, I vaguely recall seeing artefacts like these occasionally when I used to use RawTherapee; they seemed to have an algorithm that would produce exceptionally detailed output most of the time, but sometimes introduce artefacts if one wasn't careful with the choice of demosaicing algorithm for those odd images that trip it up.

Community
  • 1
PeterT
  • 621
  • 1
  • 5
  • 10
  • the exif on flickr claims it's from a raw converted with LR 6.7... – ths Nov 14 '16 at 11:45
  • Interesting, can't recall a time I've seen this with LR – PeterT Nov 14 '16 at 11:46
  • I certainly wouldn't want to imply that Lightroom is immune from such issues, but I always thought it was a bit more conservative. – PeterT Nov 14 '16 at 11:48
  • 1
    It's very odd to see this in an out-of-focus area. Demosaic moiré can only happen when there is high frequency content to give rise to the aliasing artifacts. This looks like an area of nothing but soft bokeh. The only thing I can think of is that this is perhaps outdoors (sky?) and that there was a very uniform and fine mist in the air at the focus plane (perhaps these are water droplets from a fountain also producing fine droplets or something similar?) - otherwise, this is rather odd indeed. – J... Nov 14 '16 at 12:16
  • Good point, I had rather thought that but for some reason passed over it. High frequency content could also be from noise, but that almost certainly doesn't apply here. Nonetheless something about this feels familiar, I have an inkling I might have seen something similar before. – PeterT Nov 14 '16 at 12:21
  • Very interesting answer. From what I understand, if I use another software to process the RAW image, it may render differently? – gregseth Nov 14 '16 at 13:15
  • 1
    Another software, or indeed different settings, particularly sharpness will effect the maze artefact prominence. – PeterT Nov 14 '16 at 13:32
  • Feel free to stick the RAW image online somewhere and i'll convert it with Bibble v5. –  Nov 14 '16 at 15:39
  • @J... The light is out of focus, but the dark areas of the tree are closer and the artifacts appear to be chrominance noise from the demosaicing algorithm's attempt to pull details from the shadows. Tree leaves could have the same effect as more regular types of high frequency patterns. – Michael C Nov 14 '16 at 23:00