Canon EOS 5D MkII "Black Dots" problem

UPDATE: Message from Canon dated 01/07/09 Canon have now issued firmware 1.0.7 which they say will fix these issues. The new firmware can be downloaded from http://web.canon.jp/imaging/eosd/firm-e/eos5dmk2/firmware.html or it can be accessed via the "Downloads and Drivers" tab on the support page for the EOS 5D MkII on the Canon USA website - [LINK]

UPDATE: Message from Canon dated 12/19/08

If you've been reading the popular Canon EOS web forums, you can't help but have noticed the posts concerning a "Black Dots" problem which some people are seeing in images shot with the new EOS 5D MkII. It almost seem to be "par for the course" that any new camera will exhibit some problems in its initial release and the 5D MkII appears to be no exception. However it's important to keep some perspective here. The 800% magnification image show below for example represents a section of a print that's about 16ft x 24ft when viewed on a typical monitor screen (that's 16 x 24 feet, not inches). Needless to say, on a normal size print it's going to be quite hard to see "black dots", even if you are looking for them!

EOS 5D MkII image at 800% enlargement. Black dots as bad as they get

The "black dot" problem is the presence of a small area (typically 3-6 pixels) which appears immediately to the right of some small, well focused, overexposed lights (e.g. christmas tree lights or streetlights). It seems more likely to occur at high ISO settings. You normally need to "pixel peep" quite closely to see it, and not everyone claims to see it at all, but there is no doubt that was some cameras under some conditions it is visible and it is real. In some images the dots may appear next to some lights, but not with others which are only a few pixels away, even though both lights seem equally overexposed. There may be some relation between the presence of the black dots, the size of the overexposed region and the sharpness of the light/dark edge.

A less severe (and perhaps more typical) case of "black dots"

There has been some speculation about whether it shows up mostly in JPEGs or mostly in RAW conversions and whether it depends on which RAW converter you use and if you have noise reduction turned on. However in both JPEG and RAW images produced by several different RAW converters, the black dots to appear to correlate with (typically) a small block of pixels immediately to the right of an overexposed region. The fact that they always appear immediately to the right of the bright area (with the camera held horizontally) may be related to the way in which the data is read from the sensor array (in horizontal rows, perhaps from left to right?). Perhaps some type of output signal "undershoot" as the pixel amplifer input drops rapidly from full scale (saturated) to a low value (see http://en.wikipedia.org/wiki/Overshoot#Electronics). However it's still not 100% certain if the black dots are hardware or software related. Either is still possible.

Uninterpolated data

The above image was obtained using DCRAW and using the -d option, which renders the RAW file as a grey scale image with no color or interpolation (and no other processing such as noise reduction or sharpening). As you can see the overexposed white dot (shown outlined in blue) is accompanied by 4 pixels (shown outlined in red) with an intensity of zero. This is with the black level set to the default of 1027, meaning these 4 pixels were at or below a level of 1027. They are not actually zero in the RAW file. Setting the black level at zero, these pixels can be seen to have low, but not zero values. Numbers in the 300-400 range show up (remember that 1027 is the nominal black level in the default RAW conversion), so they show up in the converted file as fully black. This is probably about as close to the real RAW data as you can get, and the black dots still show up as pixels with values lower than the black level. This doesn't positively identify the process as hardware related, but there's not a lot of software between the actual pixel output and the numbers stored in the RAW file.

Line profile across an overexposed point highlight with black dot

Whenever there is a "black dot" pixel, one of the RGB channels is always zero (in the JPEG), which means that one of the channels in the demosaiced RAW data for that pixel is at or (more often) below the black level (i.e. below the level of a pixel which has received no light at all). This suggests either a software issue where some value is being erroneously set to zero, or a hardware issue where a signal is being artificially generated by an electronic "glitch". Pixels in the image area should not be able to generate a signal lower than that of dark pixels. The dark reference pixels are usually on the outside of the array under an opaque mask, so their signal is the lowest possible signal, determined by leakage current and/or noise. Any photons at all hitting the pixel should result in a higher signal and no photons at all should result in a very similar signal. In the 5D MkII, it appears the black level (at least for the images I've looked at) is at around a pixel value of 950. Some of the "black dots" have RAW pixel values as low as 300-400.

If these black dots are a result of some sort of software problem, then a firmware update might take care of the problem. If they are hardware related, then there could be some sort of software solution, with interpolation of the pixel values in a manner similar to that used to map out dead pixels or to clone out dust specs. These dots are typically so small that interpolation of data should be very effective.

Canon have acknowledged that they are aware of the reports and their technical department is working on the issue. I'd expect to see some sort of announcement from Canon within the next week which describes the problem and what their plan for dealing with it is.

In the grand scheme of things, stuff like this shouldn't happen, but it does. It's a problem which wouldn't affect many images and which many users probably wouldn't even notice, however it's still something that needs to be addressed, and I'm sure Canon will indeed address it.

When more hard facts are known and Canon have commented, I'll update this page with the latest news.

There has been some anger expressed by some users claiming that Canon is shipping a defective product and for $2700, it should be perfect. While their frustration is understandable, their expectations are not. When working at the very limits of what's possible - such as looking at pixels next to a region of highly overexposed pixels - artifacts such as this are not totally unexpected. While $2700 is a LOT of money for me, those who think it's a fortune for a camera should look at the price of some scientific cameras. In the grand scheme of state-of-the art electronic imaging devices, $2700 isn't going to get you perfection. It should get you a camera that's capable of yielding excellent large prints for the overwhelming percentage of users under the overwhelming percentage of shooting conditions - and I think it does based on the evidence presented so far.

I also see complaints about banding and noise in the 5D MkII. Again it's not perfect. If Canon had made it 12MP and limited ISO to 800, the performance could probably have been much closer to "perfection" at all possible camera settings. However then users would complain that it didn't have enough pixels and that ISO 800 was too low. By giving users more pixels and allowing ISO settings up to 25600, Canon have opened themselves up to complaints from those who expect to get the ultimate in performance at the limits of technology.

I'm not trying to be an apologist for Canon here, I'm just just trying to put these things into a realistic perspective. The EOS 5D MkII works just fine for most applications (which don't involve imaging of greatly overxposed pinpoints of light and looking at the images at 100% magnification and higher) and at "normal" ISO settings. Yes, the 5D MkII could be better, but it's as good as it can reasonably be expected to be overall. The black dots are certainly something that Canon now need to address (and I'm sure they will), but for the majority of users, I don't think they can be considered a "deal breaker". Indeed I suspect that the vast majority of 5D MkII users whould never even have noticed them if they hadn't been pointed out. Now there are some special application areas, such as astrophotography, where they could be a problem though and for commercial photographers with really picky image editors, again it could be an issue. This is something that you'd have hoped that Canon would have found and fixed before the camera was released, but things don't always happen that way. I guess they need more pixel peeping beta testers!

If Canon said "Tough, you're just going to have to live with the black dots", then I think users would have a right to be upset. However I think the chances of that happening are very small indeed. I expect Canon to develop a fix for the problem, most likely a firmware update. Stay tuned for more news - and don't panic yet! If anyone is so disgusted with the problem that they'd like to dump their defective "black dot" EOS 5D MkII on me for $2000, I'd be glad to help them out...

UPDATE December 19th 2008

Canon has posted the following message on their website: