Topic: 5d Banding with long exposures (Read 7233 times)

Mike Chini · « **on:** August 07, 2007, 04:02:32 pm »

I've been photographing at night on and off for a while with film. Photographing in color was always a dream of mine and now, with digital, a reality. However, a problem has arisen. I've started getting really bad banding in my blue channel with long exposures (20 - 30 seconds). It seems that if I shoot a bunch of long exposures in succession, it gets worse and starts to spill over into the green and red channels. My question is this...what can I do to mitigate this. I've travelled to another city to shoot (I'm from NYC) and I have few options right now. I was wondering if I should shoot shorter exposures at higher ISO's or take longer breaks between shots (I tend to shot a lot). Of course, I'll be testing some exposures tonight but I would love to hear anyone's experiences with this problem. An example is attached...and THANKS for any help in advance!

John Sheehy · « **Reply #1 on:** August 07, 2007, 05:45:17 pm »

Quote

I've been photographing at night on and off for a while with film. Photographing in color was always a dream of mine and now, with digital, a reality. However, a problem has arisen. I've started getting really bad banding in my blue channel with long exposures (20 - 30 seconds). It seems that if I shoot a bunch of long exposures in succession, it gets worse and starts to spill over into the green and red channels. My question is this...what can I do to mitigate this. I've travelled to another city to shoot (I'm from NYC) and I have few options right now. I was wondering if I should shoot shorter exposures at higher ISO's or take longer breaks between shots (I tend to shot a lot). Of course, I'll be testing some exposures tonight but I would love to hear anyone's experiences with this problem. An example is attached...and THANKS for any help in advance!
[a href=\"index.php?act=findpost&pid=132004\"][{POST_SNAPBACK}][/a]

Banding happens during the sensor readout after exposure, so the exposure length itself is probably not a direct factor, but longer exposures may raise the temperature, and that may cause the circuitry to have more banding (1-dimensional noise). It could also be an issue with the lens. If the lens is auto-focusing during the readout of the sensor, then this can introduce additional banding besides the banding that is present in all images. Some lenses are worse than others in creating noise.

Exposure and ISO are big issues. You get the most banding when you under-expose, if only relative to your exposure. The banding is fixed; it doesn't matter what you shoot or how you expose; the banding has the same strength in the RAW data. Therefore, whatever ISO you use, you should make sure you don't under-expose. If you have a particular Av and Tv value that you want to use, never use a low ISO and under-expose it; the highest ISO that doesn't cause clipping of desired highlights will give you the least noise, including banding, with a fixed absolute exposure. High ISOs only give more noise when the ISO dictates the exposure. The lowest ISOs on a 5D (and any Canon DSLR) actually add the most noise to the absolute signal! They are potentially cleaner only because they allow a higher absolute exposure.

Oh, I just looked at your images. Are the crops from a normal landcape orientation? Most banding on Canons is horizontal. If this is truly vertical, vertical bands like that can sometimes be a "fixed pattern noise" but they are usually much, much weaker than other noises. Are the bands the same in successive images? If so, you should enable the long exposure noise reduction in the custom functions. That will add a small amount of random noise, but will remove noises that repeat from frame to frame, by taking a black exposure after the real one, and subtracting it.

Mike Chini · « **Reply #2 on:** August 07, 2007, 07:26:58 pm »

Thanks John. The banding is vertical (those were crops from a vertically-oriented frame). Everything is set to 'manual' on this camera. The lens doesn't seem to be a factor. It happens with all of my lenses. Finally, I do keep the long-exposure noise reduction set to on for long exposures, this was what happened with it!

So far, the only thing I can think of is over-heating. I'm not sure, but I think the worst banding occurs when I shoot multiple images in succession (bracket). This is important however because I love to bracket at night. I understand what you're saying about exposing to the right but at night, there is so much contrast that something has to clip badly!

Thanks for the help. I will do some testing tonight and see what's going on...I hope!

MC

Mike Chini · « **Reply #3 on:** August 07, 2007, 07:29:37 pm »

Thanks John. The banding is vertical (those were crops from a vertically-oriented frame). Everything is set to 'manual' on this camera. The lens doesn't seem to be a factor. It happens with all of my lenses. Finally, I do keep the long-exposure noise reduction set to on for long exposures, this was what happened with it!

So far, the only thing I can think of is over-heating. I'm not sure, but I think the worst banding occurs when I shoot multiple images in succession (bracket). This is important however because I love to bracket at night. I understand what you're saying about exposing to the right but at night, there is so much contrast that something has to clip badly!

Thanks for the help. I will do some testing tonight and see what's going on...I hope!

MC

John Sheehy · « **Reply #4 on:** August 07, 2007, 07:49:54 pm »

Quote

Thanks John. The banding is vertical (those were crops from a vertically-oriented frame).[a href=\"index.php?act=findpost&pid=132039\"][{POST_SNAPBACK}][/a]

That means that the banding is horizontal. That is the usual prominent banding.

Have you tried ACR? It automatically reduces banding to some degree. It's the only converter I'm aware of that recognizes this type of banding, and attempts to remove it.

Depending on the scene, I can make an offset image from RAW data to subtract banding, but no easy way to get that data back into a proper RAW converter, though. I can only use my own simple color conversion. It would be nice if RAW converters had plugin capability with SDKs available. Many of the artifacts in RAW data are quite simple, but are not addressed in their entirety by any converter.

Mike Chini · « **Reply #5 on:** August 07, 2007, 09:23:55 pm »

Yeah. That conversion was with ACR. I know what you mean about ACR being better. With DPP, the banding was nightmarish! I'll have to try other raw converters as well. How do you create an offset image to mitigate the banding? Right now, all I can figure out is to use 'surface blur.

Thanks so much for your help!

brianchapman · « **Reply #6 on:** August 07, 2007, 09:44:15 pm »

Hi Mike,

I just posted a tutorial discussing one of the techniques mentioned in the Zero Noise thread.

Here's the link to the original thread:
http://luminous-landscape.com/forum/index....showtopic=17775

If you look at post #12 you can see the beginning of banding in the noisy shadow areas.

Here's the tutorial I posted:
http://brianchapmanphotography.com/blog/?p=34

I have experienced banding with my Rebel XT and it is consistent with underexposure - and especially obvious in long exposures. I have not yet tried (maybe tonight!) the technique to fix banding noise but it's quite possible it will make a difference.

I decided to mention it because you said you generally like to bracket at night so it wouldn't necessarily be much of a difference for you during exposure time. It obviously becomes prohibitive at 8+ minute exposures but in the city you might have some good results.

If you do get a chance to try it, let me know if it helps!

Brian

Quote

Yeah. That conversion was with ACR. I know what you mean about ACR being better. With DPP, the banding was nightmarish! I'll have to try other raw converters as well. How do you create an offset image to mitigate the banding? Right now, all I can figure out is to use 'surface blur.

Thanks so much for your help!
[a href=\"index.php?act=findpost&pid=132048\"][{POST_SNAPBACK}][/a]

Bobtrips · « **Reply #7 on:** August 08, 2007, 10:43:51 am »

Just wondering...

If one is doing a session of long exposures and finding camera heat buildup to be a problem would it work to 'apply some cold'?

For example, a packet of 'blue ice' wrapped in a cloth could be held against the camera from time to time to cool it down. Or the camera could be placed in a ice chest to cool down in between shots.

Mike Chini · « **Reply #8 on:** August 08, 2007, 01:59:54 pm »

Thanks everyone.

So I think it has got to be some sort of heating problem. I went out last night and shot a bunch of shots. The first few shots were fine. Then it came back. The more frequently I shot, the worse it got. But I'm not talking about shooting constantly. What I mean is, the camera seems to need a good 20 minutes between brackets. Also, where I am right now, its pretty cool at night ~50 degrees. The strange thing however is that the banding went away when I got up to ISO 400. Also, some shots had it and some didn't. Also, the banding only seems to occur in smooth areas of shadow/lower midtones (skies at night). It doesn't seem to spill onto any other detail unless I really shoot like crazy. The frustrating part of all this is that the first 3 nights out, there wasn't one problem. I've contacted Canon and they're trying to figure out what's going on. I'll probably end up sending it in when I get back.

Any ideas for fixing the banding??? So far, I find surface blur works well and I've tried using offset layers set to dissolve and that seems to help as well.

brianchapman · « **Reply #9 on:** August 08, 2007, 02:18:25 pm »

Mike,

The banding you're seeing is pretty bad so I think fixing it in photoshop is the last resort - there must be a more fundamental cause and fix for it. Do you see it in the overexposed image of the three (or whatever #) bracketed? If so, is it reduced or just as severe? Would you be willing to send me a couple raw files (the normal and over from the bracketing?)

Brian

Mike Chini · « **Reply #10 on:** August 08, 2007, 02:32:59 pm »

Brian-

Sure thing! Let me know how to send you some raws.

MC

Mike Chini · « **Reply #11 on:** August 08, 2007, 04:56:07 pm »

Brian-

Sent you some files...

...another thought. My images also now look very monochromatic compared with what I see from other people (thenocturnes.com, Lynn Saville, Cambridge in Color). Perhaps I have more than just a banding problem?

Example attached

John Sheehy · « **Reply #12 on:** August 08, 2007, 05:46:05 pm »

Quote

Yeah. That conversion was with ACR. I know what you mean about ACR being better. With DPP, the banding was nightmarish! I'll have to try other raw converters as well. How do you create an offset image to mitigate the banding? [a href=\"index.php?act=findpost&pid=132048\"][{POST_SNAPBACK}][/a]

I didn't mention it to suggest that it is easy or pleasant to do, but it is possible with many files. The approach I take depends upon the image. The easiest to deal with are ones where there is a vertical strip that is almost black. Then, you crop the dark strip, run a highpass filter on it, resample it to 1 pixel wide, then to the width of the entire image, and subtract it or overlay its negative. With the gamma curve of converted images, though, this winds up subtracting too much from the highlights, so the highlights can wind up with some banding while you make the bands disappear from the shadows. That's where overlay works well, you can use the "blend if" functions in photoshop to only affect the shadows.

What the banding actually is, is just RAW values that are offset by random amounts on a line-by-line basis. Basically, if you can find out the offsets and subtract them, the banding is gone. This is done most optimally with RAW data. You take the green values, since green is on all horizontal and vertical lines, and get the offsets from them. The offsets are the same amount, no matter what the tonal level. They are most visible in the shadows because the signal-to-offset ratio is very low in the shadows.

Canon DSLR RAW files all contain some unexposed pixels which exhibit the banding, and this data is excluded from conversions, but has information that can be used to subtract the offsets, The only problem is that the bands of unexposed pixels are very thin, and don't offer a great sample. If Canon sacrificed a few more pixels to these bands, and had them on all four sides, I believe that banding could be almost completely eradicated in software. No one sseems to care, though. Canon creates this data, but doesn't use it to remove banding. ACR allegedly uses it, but I have had better results using the data myself, so I don't know how well adobe is doing it.

Mike Chini · « **Reply #13 on:** August 08, 2007, 05:49:29 pm »

Interesting idea. The banding I'm getting is more or less, random and 'mushy'. I've done some quick tests with the offset filter and am getting some slightly encouraging results. Of course, my goal is to mitigate the banding in the first place, but barring that, I also find that apply the surface filter to individual channels in varying amounts helps as well. Still, I think something is wrong with this camera. I will probably rent another body for the duration of my work here...

brianchapman · « **Reply #14 on:** August 08, 2007, 07:13:49 pm »

Thanks - got em.

I think this looks more like one type of light than anything - I think I can see a tiny bit of blue or white in the window or sky near the top which makes me think the mono look in the main areas is from the street lamps and could be changed with the white balance (well, that would eliminate the yellow cast but not create the appearance of multiple types of light).

Brian

Quote

Brian-

Sent you some files...

...another thought. My images also now look very monochromatic compared with what I see from other people (thenocturnes.com, Lynn Saville, Cambridge in Color). Perhaps I have more than just a banding problem?

Example attached
[a href=\"index.php?act=findpost&pid=132197\"][{POST_SNAPBACK}][/a]

Author Topic: 5d Banding with long exposures (Read 7233 times)