Jump to content

Canon 5D Mark II: Banding... c'mon, still?


rishij

Recommended Posts

<p>I have previously brought up the issue of banding in shadows for Canon digital SLRs (http://www.photo.net/canon-eos-digital-camera-forum/00RBTe). Very evident in Rebels, less evident in higher-end cameras like the 5D. But, nonetheless, there.</p>

<p>Disappointingly, it's still there in the 5D Mark II... you'll see some examples below.</p>

<p>These are more easily seen by lightening shadows using 'Fill Light' or whatever algorithm you choose. But sometimes they're visible on their own.</p>

<p>First question: Nikon CCD camera based users: do you see this?</p>

<p>Now, am I to understand that this banding results from variations in the amplifiers for each row OR from temporal variation in read-outs? And do these banding effects essentially result when these variations in gain or readout exceed the signal:noise ratio (which is much more likely to occur at low signal, i.e. shadows... OR, at high ISO where the gain for each amplifier is increased)?</p>

<p>Here are some examples from the new Canon 5D Mark II (sadly, yes):</p>

<hr />

<p>Here's an example of some <strong>horizontal</strong> banding (note, the image was significantly brightened in DPP):<br /> <img src="http://staff.washington.edu/rjsanyal/Photography/BandingInShadows.jpg" alt="Horizontal banding in shadows_ISO100" width="682" /></p>

<hr />

<p>Here's a comparison of the same area of a shot, at ISO 50 & at ISO 100, on the 5D Mark II, with Fill Light set to +100 to exaggerate the <strong>vertical</strong> banding in shadows:<br /> <img src="http://staff.washington.edu/rjsanyal/Photography/Canon5DMII_ISO50vsISO100.jpg" alt="" width="808" /> <br /> Here's the link to the <a rel="nofollow" href="http://staff.washington.edu/rjsanyal/Photography/Canon5DMII_ISO50vsISO100.jpg" target="_blank">full-size image.</a></p>

<hr />

<p>And here's the banding seen in a high ISO image (ISO 25600):<br /> <img src="http://staff.washington.edu/rjsanyal/Photography/BandingAtHighISO.jpg" alt="" width="589" /></p>

<p>One thing that confuses me though -- sometimes it seems like the direction of the banding changes from image to image, by 90 degrees (and no, not because the image was portrait vs. landscape). I'm confused...</p>

<p>Anyway, the purpose of this post, to really figure out why it's there. Berg previously suggested that it's temporal variation, and if I understand this concept correctly, I don't understand why this temporal variation has to exist if scanners like the Nikon LS-9000, which scans 1 (or 3?) lines at a time, doesn't exhibit this sort of banding in extremely dense areas of Velvia 50 slide scans, as is evidenced below:<br /> <img src="http://staff.washington.edu/rjsanyal/Photography/Banding&Dmax-LS9000Shadows.jpg" alt="LS-9000: See? No banding!" width="800" /> <br /> <a rel="nofollow" href="http://staff.washington.edu/rjsanyal/Photography/Banding&Dmax-LS9000Shadows.jpg" target="_blank">Link to Full-Size Image</a></p>

<p>And that scan was significantly lightened with the Shadows/Highlights tool in Adobe PS CS3 (Shadows: +100, tonal width=14, black clipping = 0.01%).</p>

<p>Thanks in advance for your help,<br /> Rishi</p>

Link to comment
Share on other sites

  • Replies 99
  • Created
  • Last Reply

Top Posters In This Topic

<p>Right. If you shoot at high ISO, underexpose, and use raise the fill setting... you will get banding, and noise, and a host of problems.</p>

<p>There is nothing wrong with the camera in such cases, and if you resolve those exposure and post-processing problems that you have created you won't see these "problems."</p>

Link to comment
Share on other sites

<p>Oops. I just felt I'd posted it in the wrong section since that that thread also included point & shoots and this is specifically for Canon EOS.</p>

<p>That's why I reposted.</p>

<p>Hmm... I dunno what to do about it now.<br>

Sorry,<br>

Rishi</p>

Link to comment
Share on other sites

<p>I'd assume all cameras will show banding if you look hard enough. It's almost certainly due to variations in readout noise as the array is read. I don't know how Canon readout their sensor arrays, but it's probably done on a line by line basis. A typical scheme is a parallel vertical shift register and a serial horizontal shift register, so each horizontal array of pixels is loaded into the horizontal shift register in sequence and read out serially. Tiny temporal differences in gain ("noise") can show up as banding.</p>

<p>For the Canon CMOS array I think each horizontal row is read out seperately via a multiplexing scheme (i.e. they don't have to be loaded into a shift register), so again there willo likely be very slight variation between the rows, resulting in "banding" if you look very closely at the array.</p>

<p>The only way around banding would be to read out the pixels one by one in a random fashion, which would presumably be much slower and more difficult to do. That might result in random noise rather than pattern noise.</p>

<p> </p>

Link to comment
Share on other sites

<p>BTW scanners only have one line, so they can be more uniform than sensors which have 4000 or more "lines" of pixels. Expecting every one to be identical is expecting too much. You might still see temporal changes with a scanner, so it appears the temporal component must be small.</p>
Link to comment
Share on other sites

<p>Don't sweat it, normaly the moderators delete one. This is probably the better spot for it but the other one got more interest and is older, so who knows.</p>

<p>To put your mind at rest, or start a war, I would ask the question mentioning all the IQ issues you think are there. Something like, Are the real world images comming out of the 5D MkII really affected by the fringing, black spot and banding issues, or are they just pixel peeping non isssues that only show up in excessively post processed badly exposed shots? Please only answer if you own a 5D MkII and have printed your images at (whatever size you are going to print).</p>

<p>Take care, Scott.</p>

Link to comment
Share on other sites

<p>Bob, thanks. Someone else had mentioned it's probably more a temporal component, but I wonder how much more it's due to readout noise as you mention or just differential gain between amplifiers, since each column shares one amplifier (I believe, from this: <a href="http://www.dalsa.com/shared/content/Photonics_Spectra_CCDvsCMOS_Litwiller.pdf">CCDvsCMOS</a> ). The article states that gain (for each column amplifier) was sacrificed to increase uniformity... but that can only help so much (IMHO).</p>

<p>Is it possible that since on a CCD each pixel charge is read sequentially, that any non-uniformity would be less visible (and more randomly dispersed)?<br>

Rishi</p>

Link to comment
Share on other sites

<p>You'd have to ask Canon. I don't know the details of their sensor readout scheme for the EOS 5D MkII</p>

<p>I suspect this is mostly a non-issue except for those who really like to shoot at high ISO settings and post-process the @^ out the the resulting images.</p>

<p>If you look closely enough at pretty much anything, you'll find defects. Canon never claim perfection and you shouldn't assume it. It's a camera, not a precision photometer with every pixel calibrated (and corrected) to NIST standards. Maybe if they cooled it with liquid nitrogen they could do better.</p>

<p>If you want to do some experimental science (rather than photography), take multiple images and see if the patterning is consistent. If it isn't, it's probably mostly temporal. If it is, it's probably associated with some sort of difference between rows, either due to amplifier gains or some other chip level differences in layout or connection.</p>

Link to comment
Share on other sites

<p>You're right in that perhaps I just shouldn't worry about it.</p>

<p>It's just that, to me, it's hard to stomach given that that LS-9000 scan of Velvia 50 received a +100 in the Shadows/Highlights tool (Tonal Range to which it was applied: 14 on a scale of 100), and still showed no dark non-uniformity.</p>

<p>The Imacon 848 I use, though, shows lots of dark non-uniformity, kind of like on par with what I'm seeing from certain Canon sensors... ok maybe that's a bit unfair. The Imacon (this one anyway) is probably considerably worse. Here you go:<br>

<img src="http://staff.washington.edu/rjsanyal/Photography/Banding&Dmax-ImaconShadows.jpg" alt="" width="800" /><br>

<a href="http://staff.washington.edu/rjsanyal/Photography/Banding&Dmax-ImaconShadows.jpg">Link to Full-Size Image</a></p>

<p>Again, same Shadows/Highlights adjustment to the Imacon scan. And apparently the Imacon's CCD is peltier-element actively cooled.</p>

<p>Some really cool results of cooling a Canon DSLR sensor here: <a href="http://www.pbase.com/terrylovejoy/image/65814409">http://www.pbase.com/terrylovejoy/image/65814409</a></p>

<p>Still don't know if it'd get rid of non-uniform noise in the form of banding due to differential gain of amps though. Not that active cooling is practical (or necessary?) in DSLRs for practical use anyway...</p>

<p>Cheers,<br /> Rishi</p>

Link to comment
Share on other sites

<p>Though I'm hardcore Canon, I have a theory that Nikon cameras don't show this banding. Not their CCD cameras anyway.</p>

<p>I'd love for someone from the Canon camp to challenge me. I haven't seen this banding in the limited set of Nikon CCD camera images I've attempted to bastardize via Fill Light.</p>

<p>I don't have a Nikon. Open to debate. Maybe I should post this over in Nikon forums to see what they say.</p>

<p>Rishi</p>

Link to comment
Share on other sites

<p><tt><em>"Maybe if they cooled it with liquid nitrogen they could do better."</em></tt><br />    Actually, doing this would lead to a disaster: at 77K, the microlens array will crack and fall apart, and the chip will fail completely due to carrier freeze-out. Cooling mainly helps to reduce dark current which is very well controlled in modern image sensors.</p>
Link to comment
Share on other sites

<p>OK, so if dark current is well controlled, why in the name of $%&* is column amp variation not?</p>

<p>I know $2700 for a body may be nothing for you folk in the industry, but it's a lot for a hobbyist. It was hard enough for me to shell out $3000 for the 'holy trinity' of L-series lenses, and $750 for a Gitzo tripod + Manfrotto head capable of holding a 35mm camera still enough for longer exposure shots.</p>

<p>Before I shell out thousands more for a digital body, I want to know that it's worth more than the $18/roll of Velvia I pay. Wait a minute, that doesn't account for the countless # of hours I spend on a $12k Imacon that can't even function correctly...</p>

<p>So I guess the moral of the story is: current consumer technology clearly does not satisfy me. Go figure [i'm a self-proclaimed pixel peeper].</p>

<p>-Rishi</p>

Link to comment
Share on other sites

<p>Rishi,<br>

Yes, banding noise is the result of temporal and sometimes permanent variations in the row/column sense amplifiers, at high ISO since the gain is increased this non-uniformity is more visible in the form of vertical or horizontal stripes.  This is also true about deep shadows, there is very little signal in the shadows so you're just exploring the noise floor of the sense amps just above the dark current. There are a few points you need to consider about 5d MKII though, the ISO 12,800 and 25,600 are not real ISO's they are derived from 3200 values by sw multiplication of intensity levels by the corresponding factor. This amplifies any non-uniformity and noise. The ISO 50 image you posted doesn't show any banding, all I see is posterization, ISO 100 shows some but I think this is due to the fact that it has been heavily raised ve it's shadow levels in post processing.<br>

Also regarding scanners, a scanner is much slower than a 5DMKII which can shoot at 4fps, thus temporal fluctuations due to white noise are averaged out, so this is expected.  <br>

Banding is not just an issue with CMOS, in fact the most acute from of banding was observed in Nikon D200 which had a CCD and a particular issue with the readout circuit, Nikon provided a hardware fix for that. Also no Nikon camera currently in production uses a CCD, they are all CMOS.</p>

<p> </p>

Link to comment
Share on other sites

<p>" Actually, doing this would lead to a disaster: at 77K, the micro lens array will crack and fall apart, and the

chip will fail completely due to carrier freeze-out. Cooling mainly helps to reduce dark current which is very well contro

lled in modern image sensors."

<br>

Brag, the patterned oxide layer (microlens) is epitaxially grown and will survive all the way down to 100mK

, If temperature is reduced gradually. A Silicon transistor will perform much better at lower temperature due to red

uced phonon scattering. You can increase the clock speed of your CPU by a factor of 2 or more if you cool it down to

LN2 tempratures. A silicon photo detector will also perform much better at low temprature due to much re

duced SRH recombination of optically generated EHPs (silicon being indirect gap material) and trap&n

bsp;assisted recombination as well as reduction in dark current and Johnson noise. At 4K sensor will operate

close to quantum noise limit, although you might see some weird coherence effects.  I routine

ly test IR detectors at 4K in our lab to beat noise. A moderately doped (>1E18 cm^-3) silicon wafer will co

nduct nicely at 77K despite the fact that it has lo

wer carri

er conce

Link to comment
Share on other sites

<p>I never understand you people here... while most people are trying to reduce noise in their images, you guys are trying to exagerate it as much as possible.<br>

From all i've read, the 5D is the best camera for noise ever made... are you guys still complaining? Just go take some damn pictures and enjoy it.</p>

Link to comment
Share on other sites

<p>Well said David....I find these forums amazing..inhabited by pixel peepers and 'photographers' with way too much spare time on their hands...why aren't they out taking pictures? <br>

Way too much fixation on one aspect of the photographic chain and way too much time looking at images at 100% instead of normal viewing distance.<br>

Get a life folks!</p>

Link to comment
Share on other sites

<p>Well said David....I find these forums amazing..inhabited by pixel peepers and 'photographers' with way too much spare time on their hands...why aren't they out taking pictures? <br>

Way too much fixation on one aspect of the photographic chain and way too much time looking at images at 100% instead of normal viewing distance.<br>

Get a life folks!</p>

Link to comment
Share on other sites

<p>Again, well said David!<br>

If you deliberately underexpose a high ISO image and then brighten it up in PP you're asking for banding. Why even bother doing that? Would you underexpose your work in really life? The answer is likely "NO". If you expose your images correctly and there's still banding, now we have a serious problem. </p>

Link to comment
Share on other sites

<p>Arash - Microlenses are typically formed in patternable organic material like photoresist. I've never heard of any epitaxial layer used for a non-crystalline material, e.g., silicon dioxide cannot be epitaxially grown.<br />     Sure, a transistor will work at cryogenic temperature, but the photodiode region in the pixel has a lightly doped region where the carriers will freeze out at low temperature. 1E18 is actually a highly doped material. CMOS sensors are typically fabricated on 5E14 doped epi silicon. IR detectors hybridized to silicon readout multiplexers work better when cooled but commercial monolithic silicon detectors won't. Modern sensors have pinned photodiodes with very low surface recombination velocities, yielding high sensitivity and low dark current.</p>
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...