michael49 wrote:
Here are three 5d classic shots in which I increased the brightness in LR by 80-100....(I can't post crops right now, but you can definitely see increased noise in the shadows, but no banding)....
Do you expect that banding would show with the 5dII with these types of shots?
Thing is that the brightness slider in LR is a mid tone adjustment, it just isn't pushing the deep shadows anywhere near as aggressively as the shadows or curves or fill light tools in LR or ACR. I'm also interested in your question though for precisely that reason.
From a quick play around with these files (I'l try and post one up later with some crops) to create banding I need to have really fallen off the end of the histogram in the shadows, with no detail in the area of banding (Backgrounds, sky, etc). If there is any detail the banding is much much harder to notice, unless i'm literally dragging it up from blackness.
I've just been applying Shadow/Highlight tool of +50% for the shadows to notice any effect. This is WAY more than I would ever ever do to an image in PP. I'm not saying I wouldn't like more dynamic range, but for me to reproduce these effects I have to go to extremes I otherwise wouldn't.
Edit: These are being processed in C1 1.5, which from the quickest of quick comparisons, seem to handle the effect (likely by dragging more detail from the shadows) better than ACR.
Daan B wrote:
3) The quickest way (for me) to reveal any kind of banding is pushing the LR2.5 fill light slider to +100 (maximum). Even though I would never use such a drastic measure for real... only for testing purposes.
Well, yes. Of course.
Pushing fill to maximum is going to make any shot look pretty awful, especially if you do this to compensate for underexposure. In CS4 I avoid fill higher than about 10 except in extreme cases.
This is a 100% crop taken from a very underexposed frame - see the small histogram embedded in the lower part of the image.
As the embedded text explains you see five versions of this section. The lower half is unaltered - no changes to color balance, brightness, etc. during conversion from RAW to 16-bit CS4 file. In the upper section successive additive +100 brightness adjustments were made to this virtually black material. You can easily see how radically the image was boosted by noting that in what was an almost black area there are areas getting close to white in the +400 upper section, and than the tree branches to the right that were slightly underexposed are badly blown out by the top. (The blue tint is from illumination of this deep shadow area by open sky only.)
If I look closely I'm pretty certain I can see some banding at +400 and at 100% magnification I think I can maybe see some at +300. But in neither case would this be visible in a fairly large print - and it is the result of trying to rescue a poor exposure by extreme measures, not the result of the camera malfunctioning.
gdanmitchell wrote:
Pushing fill to maximum is going to make any shot look pretty awful, especially if you do this to compensate for underexposure. In CS4 I avoid fill higher than about 10 except in extreme cases.
Agreed... I only took this drastic measure to determine wether there was any banding visible at all. But I could have used other tools (like levels, curves, shadows, brightness, etc) as well for this purpose.
If I look closely I'm pretty certain I can see some banding at +400 and at 100% magnification I think I can maybe see some at +300. But in neither case would this be visible in a fairly large print - and it is the result of trying to rescue a poor exposure by extreme measures, not the result of the camera malfunctioning.
Yes, the combination of underexposure and pushing seems to result in banding. But not all of the time... not even after a drastic +100 fill light adjustment. "Why is that?" is the question I still like to see answered
AdrianRogers wrote:
From a quick play around with these files (I'l try and post one up later with some crops) to create banding I need to have really fallen off the end of the histogram in the shadows, with no detail in the area of banding (Backgrounds, sky, etc). If there is any detail the banding is much much harder to notice, unless i'm literally dragging it up from blackness.
Interesting point. Although I have seen banding in underexposed images that still had a lot of detail in the shadows. But I am not sure if I haven't seen banding in images with no detail left in the shadows...
One thing not mentioned yet is the effect of Highlight Tone Priority. I think the nature of what HTP does makes noise (and banding) issues more noticeable. Sometimes I wonder if many of the examples we see of banding have HTP on or off.
Okay here is a picture followed by some crops. Sorry for the size of the crops, but they're 100% and I wanted to show the whole eye section to give you an idea of what was happening.. Click the thumbnails for full size.
Original image, no colour adjustments, no exposure, no sharpening, processed as 16bit TIFF with Capture One Pro.
Image with +50% default (Amount 50%, Tonal width 50%, Radius 30%) Shadow recovery with the Shadow/Highlight Tool in Photoshop CS4.
100% crop of the left eye from the original image
100% crop of the Shadow recovery image
How the final image actually ended up, which pretty much blacks out that bit anyway Go figure!
As you can see, the banding only really begins to present itself in the deepest darkest section of the eye (And even there I can only see a couple of more noticable horizontal lines). Skin detail in the normal shadows is held very well, and there is even a bit of detail in the really dark bags. As you can also see, 50% Shadow recovery is something a sane person would never do to the original image.
I appreciate however, this does not show the most testing scenario of dynamic range. But I do still feel that I need to really fall off the end of the histogram (and subsequently drag it back up) to really produce this banding. Unless it's on something with no detail at all, whereby it then seems easier to do.
The following isn't really for any sort of real comparison (brainiac would crucify me!), but I thought i'de include them for a little referance. Referance to what i'm not sure, as they were just off camera jpegs whilst bored waiting for someone to turn up, but these files are from a 1DSMK2, just to give a little hint at how they performed in the skin shadows at 100% (Also remember, there is no detail in this 'skin' to mask any noise, it's perfectly smooth)
Full untouched image
Full image +50% Shadow recovery in CS4
Crop of jaw/neck area where it goes into the deepest shadows, untouched
Same crop with +50% Shadow recovery in CS4
Hope these samples are to (someones) satisfaction I've got a few more but I really must be getting on with some work! Needless to say, whilst I wouldn't say no to a little more low ISO optimisation in lieu of some of the amazing high ISO performance of this camera, i'm certainly not going to send it back for the banding.
AdrianRogers wrote:
Okay here is a picture followed by some crops. Sorry for the size of the crops, but they're 100% and I wanted to show the whole eye section to give you an idea of what was happening.. Click the thumbnails for full size.
Original image, no colour adjustments, no exposure, no sharpening, processed as 16bit TIFF with Capture One Pro.
As you can see, the banding only really begins to present itself in the deepest darkest section of the eye (And even there I can only see a couple of more noticable horizontal lines). Skin detail in the normal shadows is held very well, and there is even a bit of detail in the really dark bags. As you can also see, 50% Shadow recovery is something a sane person would never do to the original image.
I appreciate however, this does not show the most testing scenario of dynamic range. But I do still feel that I need to really fall off the end of the histogram (and subsequently drag it back up) to really produce this banding. Unless it's on something with no detail at all, whereby it then seems easier to do.
... ...Show more →
Great thread Daan, thanks for putting some effort into this.
I'm going to dig through some old files, but I would wear that I had banding issues with my old 30d, and only when abusing fill. I too would like to see banding straight from camera. If it doesn't exist, I'm not sure there really is a problem.
vtotter wrote:
One thing not mentioned yet is the effect of Highlight Tone Priority. I think the nature of what HTP does makes noise (and banding) issues more noticeable. Sometimes I wonder if many of the examples we see of banding have HTP on or off.
michael49 wrote:
Nothing there that would bother me.
+1
But like Adrian already mentioned, this sample doesn't exceeed the DR of the sensor excessively... in fact, it stays well within the boundaries of the histogram. So detail is recorded in the shadows as wel as in the highlights.
Navyblue wrote:
I wonder if it is related to the amount of details? In your first example there are lots of fine detail, and there are less of it in the second example. Either the pattern is drowned by the fine detail, or when the neighbouring pixels are recording very different value somehow the circuit is less prone to interference (assuming it is caused by electronic interference)..
That's exactly what I'm seeing. In the first image the details "drown out" the pattern noise, but not entirely. I still see it somewhat.
musclepics wrote:
That's exactly what I'm seeing. In the first image the details "drown out" the pattern noise, but not entirely. I still see it somewhat.
Still, in the underexposed cat shot there is banding (after pushing). In the overexposed cat shot there is no banding. Crops from both pics show the same background (with not many details - wooden planks).
But I will see if I can find an underexposed shot with much less detail that doesn't show banding after pushing the shadows (to determine if the amount of details is somehow related to the banding)
Uses of the terms underexposure and overexposure on this thread are basically meaningless. Different scenes are of different contrast. Typically a photographer must underexpose darker parts of a scene and overexpose brighter parts. If the contrast in the scene exceeds the DR of the camera, then over and under exposure for the scene as a whole, are not defined. A photographer decides whether to blow highlights, or whether to raze shadows. It is arguable that optimal exposure on a 5D2 is the exposure which retains the very slightest detail in the brightest part of the image, so that after intelligent processing there are no parts of the image where highlight compression occurs due to saturation of the photosites. In order to shoot this way, a landscape photographer would be advised to use the neutral picture style with minimum contrast and saturation so that her histograms and review jpegs show what the sensor is capturing, notwithstanding the gamut limits of the camera's monitor.
The limits of the camera's dynamic range are felt separately in each of the RGB channels. A scene can have enormous contrast in the red channel and almost none in the blue. The same goes for brightness/exposure levels. That is why light temperature can have a dramatic effect on the noise levels found in each separate channel, and why parts of an image that contain brightness in only one primary colour can show much stronger banding than greyer areas. Tungsten bulbs emit much less blue light, so when images are balanced to tungsten it is often the blue channel that becomes noisy and bandy.
Consequently it is pointless to regard 'fill light' type processing as excessive. It is just another way of maximising the camera's capabilities. Some people like large areas of blackness in their images. Other prefer not to use a camera as if it's a photocopier. I have seen banding on 5D2 shots which were exposed to preserve highlights, and then processed to maximise dynamic range by setting contrast to minimum, i.e. opening up the shadows. The visibility of banding is certainly affected by how much detail lies in the shadow areas of the image, how deep the shadows are in brightness (i.e. how far they are in exposure from the brightest parts of the image, assuming the highlights are not blown), and colour balance.
The problem with the 5D2 is that it has large scale banding in its noise footprint which is clearly visible at web image sizes. This kind of banding noise is virtually impossible to fix and it means that shadows can not be retrieved as successfully as they could be if it weren't there. Canon has clearly tackled this limitation in subsequent cameras, namely 500D, 7D and 1D4, so we can expect the 5D3 to have much less banding, and probably be a much better camera for coping with high contrasts scenes. Due to 5D2's banding, and the fact that it has been fixed in subsequent Canon cameras, we can also expect the 1D4 to exceed 5D2 in dynamic range and credible tone. It will be a great shame if those who don't like 1 series cameras are made to wait 2 more years for a light full-frame camera that lacks this flaw.
So what does it take? I have seen banding at 100 iso in an azure sky that was a stop underexposed. I have seen it in shadows of a street scene at iso 100 when the sky was only just not blown. It's all over the place, but in most cases it is slight and only subtly undermines the sense of reality in the image. I wonder if the horizontal line in the crop of the woodland scene at the top of this thread is banding. I think it's a feature of 5D2 files at all isos whenever you use low contrast settings in order to cope with high contrast scenes. The effect is usually subtle but it's often there. It might take a trained eye to spot it, but it only takes an average eye to be subliminally effected by it.
It seems to me that compression of colour kicks in earlier than people suppose. With slide film one had to ensure that colours weren't too bright if you wanted rich strong colour. Digital seems to be the same, in that saturating the highly exposed colours enough leads to excessive saturation in the shadows. Of course, it's possible to tie saturation to exposure in Photoshop, but it's not trivial and there may be tonal penalties for doing so. It may be the case that high-key colour suffers when you expose to the right trying to boost DR.
brainiac wrote:
I am curious as to whether it is possible to get the same highlight qualities from this file[...]
I am a little busy atm... but you can check for yourself pretty easily. Photograph a high contrast scene in a fixed position. One with exposure for the highlights (no highlight clipping) and one with overexposure by 1 stop. Bring the overexposed file back with -1 and compare colors. You can also open up shadows first and then compare. It should be pretty evident
Question: Is anyone calling their regional Canon service centers about this? Has anyone posed the question directly to Chuck Westfall at digitaljournalist.org?
I've called a Canon Service Center a while ago...I've even given the rep the link to one of the banding articles in this forum. She has accessed the link.
RDKirk wrote:
Question: Is anyone calling their regional Canon service centers about this? Has anyone posed the question directly to Chuck Westfall at digitaljournalist.org?
Go figure!
