I’ve had the 5D Mark III for a week today, and one of it’s obvious great features if the ability to practically shoot in the dark. The high ISO performance of the 5D Mark II was amazing, and this camera surpasses that by a good few stops. However, when taking some test shots over the last week, on reviewing I had noticed some strange “banding” on the LCD screen on the back of the camera at moderately high ISO:
I wanted to better understand why this banding came about as it’s not a particularly nice artefact to have on images… is it even on the real image or is it just something you see on the LCD? I noticed it at ISO 6400 which – while high – is supposed to be comfortably inside the usable range of the 5D Mark III.
Below are two examples of 100% crops from 3 images.
And the second is the bottom section of a sky shot at dusk:
In the top images (the extracted JPEG), you can clearly see the banding as horizontal streaks of colour – mostly red (note that I haven’t changed these images in any way from the output – you can see the colour streaks more clearly on the LCD as the contrast is different to a PC monitor). In the second image, there is a lot more colour noise in the image. What you can tell if you compare the images on top of each other is that all of the banding in the top image is present in the middle image too – there’s just a lot more noise to mask it. The third image with the default noise reductions settings in DPP doesn’t show much noise, but doesn’t keep much pixel level detail either. I’d say the noise reduction is probably a bit heavy here.
Comparing the file sizes gives you an idea of how much detail is in the images:
Test1: LCD Preview: 2.96MB, DPP (NR OFF): 13.2 and DPP (with NR): 7.68
Test 2: LCD Preview: 3.29MB, DPP (NR OFF): 14.7MB and DPP (with NR): 7.50MB
Note that all images are the same maximum resolution of the sensor – 5760×3840 pixels.
I think what happens is that the LCD preview is both heavily and quickly compressed. This means you get a fairly quick preview, but the JPEG parameters will not be optimal for the rendering. This also applies the in-camera settings for picture style, noise reduction etc. so the image is softer than the raw (no noise reduction) image. The noise reduction works by reducing randomness in the image, but the banding artefacts are not completely random. In order to appear as banding, some pixels must be dependant on their horizontal neighbour.
The 2 final converted images from DPP both mask the banding, as does a conversion in Lightroom (using ACR6.6 Beta from a DNG).
I don’t have much in the way of conclusions here, other than to say there definitely seems to be some streaking on my 5D Mark III sensor at ISO 6400. Whether this is an issue or not, I don’t know. It could be that it’s there in the 5D2, 7D, 1D Mark IV, Nikon D3s etc and it’s just an artefact of general sensor design and read electronics and no-one worries about it. It may also be that my 5D Mark III is faulty. I plan on doing some further testing with the 5D Mark II, 7D and 1D Mark IV and Nikon D7000 to see how they behave under similar conditions to see if there is any evidence of an issue here. It’s worth noting that this is very unlikely to affect real-world images.