5d mark II strange black dots?

[www.danahoff.com]

<p>Canon says:<br>

"Thank you for your inquiry. We value you as a Canon customer and appreciate the opportunity to assist you. We are sorry to hear that you are reconsidering the purchase of your 5D Mark II. </p>

<p >It has been reported that some users are experiencing a phenomenon where small black dots appear when small, bright, point light sources are included in a photograph. Canon Inc. recognizes this issue, and is preparing a countermeasure. We will provide further details as they become available.</p>

<p > </p>

<p >We certainly apologize for any inconvenience this causes you. </p>

<p > </p>

<p >We hope this information is helpful to you. </p>

<p > </p>

<p >Thank you for choosing Canon."</p>

[dennis_knoxville]

<p>Bob Atkins<br>

Note above the post from Berg Na on Dec 6 on the <a href="http://www.appliedcolorscience.com/black_sun.htm">Black Sun effect on CMOS sensors</a> for one explanation.</p>

[rishij]

<p>I don't like that 'Black Sun' explanation b/c if it were true then at ISO 50 you'd see MORE black dots because ISO 50 essentially overexposes the image by 1 stop, then 'pull-processes' it back to 0 EV in order to bump up detail in the shadows.</p>

<p>Think about it, if it really were the Black Sun effect, then pixels that were near saturated would be MORE LIKELY to be saturated by being overexposed by 1 stop; that is, the black dots effect would be MORE prominent at ISO 50.</p>

<p>Yiannis' explanation seems to make sense, but he hasn't answered my question yet about amp gain, etc. So, I'm waiting to hear back from him.</p>

<p>Rishi</p>

[xavier_henri]

<p>Rishi, I'll try to explain... The gain is fixed by the ISO setting, what Yiannis is really referring to is <a href="http://en.wikipedia.org/wiki/Overshoot">overshoot</a> (look at the Electronics section).<br /> What happens is that when the signal goes from saturation to a low level (at the row amplifier), it has a negative overshoot period. If the sampling is done too early, the signal has not been stabilized and is very low.<br /> <img src="http://upload.wikimedia.org/wikipedia/commons/f/ff/Clock_signal.gif" alt="Overshoot chart from Wikipedia" width="560" height="420" /><br /> If importing the image is not done via DMA (Direct Memory Access), there might be a very good firmware solution that rereads the pixel if it is detected as negative overshoot (below the standard dark value). Otherwise, they might need to slow down the reading if that's possible or do some interpolation to replace the faulty pixels...<br /> <br />Xavier (with some time in VLSI Design engineering more than ten years ago ;-)</p>

[rishij]

<p>Thanks Xavier. I love this community :)</p>

<p>Your explanation makes things make more sense. It's a fixed gain across a row of photocells, due to the fixed gain of the row amp. Yes? I think I was making it more complicated.</p>

<p>So this means that the row amplifiers <em>cannot</em> apply non-linear gain, that is, higher gain for photocells exhibiting lower charge and lower gain for photocells exhibiting higher charge?</p>

<p>Because that would require reading the charge of the photocell, and then applying an appropriate gain. That's not possible, is it?</p>

<p>Also, would variations in the exact gain applied by these row amps give rise to banding (<a href="00Rjyp">http://www.photo.net/canon-eos-digital-camera-forum/00Rjyp</a> )?</p>

<p>TIA,<br>

Rishi</p>

[rishij]

<p>Plus, a picture is worth a thousand words. Thanks for that graph!</p>

[rishij]

<p>Wait a minute, Xavier, is the Y-axis in your graph the voltage the amplifier is getting from a given photocell?</p>

[rishij]

<p>Nevermind, that's just a generic graph explaining overshoot.</p>

<p>So now I'm confused again, but most likely b/c I don't understand the fundamentals of how the charge is read & amplified. Isn't the same amplifier gain being applied to all pixels (determined by the ISO)? So then a bright saturated pixel will receive the same amount of gain as a dark pixel, correct?</p>

<p>It's just that if you're reading the dark pixel during a 'negative overshoot' period of the amplifier, then you may darken the pixel and get a blacker pixel. If you're reading the sensor from left to right, then this would show up immediately to the right of a brighter pixel because a negative overshoot would be more obvious next to a bright signal (pixel), correct?<br /> <br /> But, then, theoretically this negative overshoot could occur anywhere in the entire image, not just next to saturated pixels, right? Or is it because the amp has to 'recover' from a high voltage output when a saturated signal comes in from a saturated pixel, causing a negative overshoot, which then affects its reading/gain application to the dark pixel immediately next?</p>

<p>OK, I think I just rationalized it to myself in that last paragraph. Let me know if my thinking is wrong.<br>

Cheers,<br>

Rishi</p>

[rishij]

<p>Oh boy, take a look at 1:01 onwards in this video... pay close attention to the center tip of the candle flame. A few times, huge black/white dots that look like huge pixels (just 1 at a time) will pop up... my guess is these were black dots that the video encoder turned into a big black macroblock during compression.</p>

<p><a href="http://uk.youtube.com/watch?v=CdaybWRRDWI">http://uk.youtube.com/watch?v=CdaybWRRDWI</a></p>

<p>Pretty hideous. Really hoping Canon can do something soon about this. The video linked above is inspiring, even though I'm a photographer, not a videographer.</p>

[xavier_henri]

<p>Rishi,<br>

Essentially, your last paragraph is correct: there is a negative overshoot because the voltage is coming from far off. It seems to me from the pictures I have seen that it happens only on small saturated dots. Which lends me to think that the conditions in which it happens is a small area of the dark/saturated/dark type. This would translate at the row amplifier to a positive overshoot (that we don't see) followed with the acquired "momentum" by an important negative overshoot.<br>

 

<p>That's one possible explanation which I like, but there are others possible, including a bug in software signal processing...</p>

</p>

[peter_wang6]

<p>To the administrator(s):<br>

This page is still automatically linking to the 6887.com domain via ftp. This sort of behavior should be disabled by default in the forums, as it poses a security risk. Please correct this issue promptly as this opens your site to malware via redirection attacks.</p>

[bobatkins]

<p>Peter - I think I've fixed that issue here. Let me know if you see it again (typo by poster).<br>

I have no control over what is and isn't allowed in the forums, but you are right that it's an issue we need to look at. Since I'm no expert on this, you might want to send a note with your concerns to Jin Choi (<a href="http://www.photo.net/photodb/user?user_id=19867">http://www.photo.net/photodb/user?user_id=19867</a>), who is our systems admin contact. I'll send him a note too. Thanks. </p>

[rishij]

<p>Xavier,</p>

<p>Your hypothesis is the ONLY one that makes sense so far. Again, because with the 'Black Sun' the effect would disappear at high ISOs. And, your theory predicts that the effect would be subdued at ISO 50, since a lower gain is applied by the amps.</p>

<p>Perhaps the sensor is read too fast because of the implementation of HD movie mode in this camera?</p>

<p>Rishi</p>

[jamesg]

<p>I received my 5DMII yesterday and immediately started testing. I too have the black dot issue... as expected. But a couple of things that I found:<br>

You can see it on the LCD screen at full zoom. I think this rules out RAW processing.<br>

I also tested my 1DsMIII and it did not have the problem. This confirms what others have said, it doesn't seem to be the sensor. <br>

So, I'm thinking that we need a solid test performed on a 50D to see if it is the DIGIC 4. <br>

I have listed out the exact test I performed here: <a href="http://douglasgoodinphotography.blogspot.com/">http://douglasgoodinphotography.blogspot.com/</a><br>

Feel free to wail on me if I have been reduntant or completely wrong! :^) <br>

Thanks,<br>

Doug</p>

[jamesg]

<p>Xaivier,<br>

Excellent explanation of overshoot as it applies to the MarkIII problem. (I was an ASIC designer in a previous life :^) <br>

But, wouldn't the 1DsMIII have the same issue? I thought that they both used the same sensor... and therefore the same amplifiers etc. Am I mistaken? <br>

Thanks </p>

[xavier_henri]

<p>James,<br>

I believe that they are probably reading the sensor data much faster on the 5D² to catch up with the video 30 fps...<br>

So reading faster means sampling sooner, that is before signal stabilization!<br>

Oh, and it appears that they are reading 4 pixels at a time, which explains the black dots size...<br>

Search for "four channels" <a href="http://www.canon.com/camera-museum/tech/report/200810/report.html">here</a>.</p>

[jamesg]

<p>Xavier,<br>

Yep, that's it. You nailed it! Thanks for taking the time to point me to the improvements section of the 5DMII tech report. And with that we can ignore my request for a test on the 50D. :^) </p>

 

[rishij]

<blockquote>

<p>I believe that they are probably reading the sensor data much faster on the 5D² to catch up with the video 30 fps...</p>

</blockquote>

<p>Woohoo! Looks like I may have been right, according to Xavier, when I said:</p>

<blockquote>

<p>Perhaps the sensor is read too fast because of the implementation of HD movie mode in this camera?</p>

</blockquote>

<p>If so, can't they fix this in firmware? Like for photos readout more slowly, but for video keep the readout fast? Also, Xavier, Canon has probably figured out the cause already, right? Otherwise perhaps you should e-mail them? I'd rather they not just put some software fix to try to 'clone' out black dots...</p>

<p>Rishi</p>

[xavier_henri]

<p>I'm sure Canon knows what's really going on...</p>

[steve_may1]

<P>This might be interesting for some of you technically minded people. On TV last night I watched a short programme about someone who makes very small models and puts them inside the eye of a needle....a helicopter, a person, a flower. They fixed a TV camera to a microscope to see the models in detail (unbelievable!), The needle was backlit, and there along a high contrast edge was a line of black dots, exactly the same as I have seen exhibited by the 5D Mark 2

[rishij]

<p>Where's Vijay?! I'm sure he'd have some explanation invovling 'sentient black dots'!<br>

:)<br>

-Rishi</p>

[bernardwest]

<blockquote>

<p>And, your theory predicts that the effect would be subdued at ISO 50, since a lower gain is applied by the amps.</p>

 

</blockquote>

<p>Rishi, I don't think that is correct. From what I can tell ISO 50 must be implemented in the raw software, not in hardware. When you compare an iso 50 to iso 100 with same exposure (same Tv, Av), the undemosaiced raw data is identical. ie. they are both iso 100 shots. The 'L' iso shot should receive the same amplification as the 100 iso shot (which is the native iso). By the way, I compared this on a 5DmkI. I am assuming 'L' iso is implemented the same way on the mkII.</p>

[yakim_peled1]

<p>

<p>What I like about this problem is that Canon acknowledged it straight away. It is an indication that they have concluded that the way they treated the former problem was a mistake. That gives me hope.</p>

<p>Happy shooting,<br>

Yakim.</p>

</p>

[allen_edwards2]

<p>

<p >This is a very interesting discussion on the 5D2, a camera I am considering. As a long time circuit designer, I suspect this is not overshoot but rather the amplifier having very low gain as it recovers from saturation. If an amplifier is not designed carefully, its gain will go way down when you saturate it. Entire sections of amplification may become unbiased, for example. In this case, the next values would read some very low level. After a period of time, the amplifier bias levels return to normal and the gain goes back up. As this is not affected by ISO settings, the offending stages may be near the output, after the gain control.</p>

<p >They may be able to slow things down and allow the amplifier time to stabilize. On the other hand, if they need this fast sample rate, the solution would be a new amplifier design and not firmware. There are things you can do in a circuit design to keep things biased under high input level conditions. Basically, you put circuits in to lower the gain of the amplifier gracefully so that nothing gets unbiased. In this case, the recovery can be very fast. The circuit design time would not be fast, however.</p>

<p > </p>

</p>

[rishij]

<blockquote>

<p>And, your theory predicts that the effect would be subdued at ISO 50, since a lower gain is applied by the amps.</p>

</blockquote>

<p>Yeah I'm not so sure I believe my own statement here since ISO 50 is actually an ISO 100 shot overexposed by 1EV (an ISO 50 shot looks just like an ISO 100 shot except that the brights get brighter, and sometimes blow out... hence the limited dynamic range at ISO 50). So, even though the gain may be different (Bernie, it has to be lower to get the same RAW data from a +1EV overexposure, yes?), you'd have more saturated pixels... so I don't particularly see how this should decrease the black dot effect... yet it does.</p>

<p>Rishi</p>