[dosdan]

Compare 3 top-end cameras:

Camera	Format	MPixel	Pixel Pitch
Pentax K-5	APS-C	16Mp	4.75 µm
Canon 1D IV	APS-H	15Mp	5.7 µm
Nikon D3x	FF	24Mp	5.9 µm

DxOMark - Compare sensors

While the price difference is enormous, what is also noticeable is the relatively weak low ISO performance of the Canon (the 7D is similar).

The K-5 has the smallest pixel pitch. When the Screen tab (to compare sensel-to-sensel performance, regardless of Mp) is used in the DxO SNR 18% chart, the sensel performance is close for all three. If the Print tab is used (normalised to 8Mp output size so that the better noise performance of extra Mp are accounted for when producing a fixed size 8"x12" printout), the Nixon D3x, with its 24Mp, powers ahead. But SNR 18% is for the typical exposure level in a well exposed shot, so this is mid-tone noise.

When looking at the DR, where the saturation level is compared to the noise floor, the situation changes. Under the Screen tab, the K-5 sensel, although having the smallest sensel pitch, is the best performer at ISO400 and below, due to its new technology. The Nikon, under the Print tab, is now similar to K-5, due to more Mp. The Canon is superior to both from about ISO 2500 and above, but at low ISO, its DR performance collapses.

Roger Clark mentions how the intrinsic read noise of the 1D IV at high ISO has set a new standard for performance: 1.7 electrons. With its full well (saturation) capacity of 55,6000 electrons per sensel, he works out the sensor's DR as:

Sensor dynamic range = 55600/1.7 = 32700 = 15.0 stops (log32700/log2).

But this in not the camera's low-ISO DR, because the read noise rises to 22.2 electrons @ ISO100 with the DR now: 55600/22.2 = 2504 = 11.3 stops. This is similar to DxO's DR value under the Screen tab: 11.46 stops @ ISO100 (indicated).

In his Conclusion, he states:

Dynamic range is still limited to a little over 11-stops, apparently by downstream electronics that must process the data extremely fast (at 10 frames per second; that is over 160 megapixels/second). I would like to see a camera option that used a slow 16-bit A/D converter and low noise amplifiers to deliver data with the full capability of the sensor, which is 15 stops. Even if it took 10 seconds to read out the sensor (longer readout times are used in scientific applications), there are situations where high dynamic range imaging would benefit.

Clarkvision.com: Canon 1D Mark IV Sensor Analysis

Now compare the DR of 3 other Canon models, under the DxO Screen tab:
5D II
60D
7D

At the single sensel level, all these cameras' DR only reaches about 11 stops @ISO100. So I don't think Roger Clark's explanation of the relatively poor low ISO DR of the 1D IV is the reason.

DxOMark - Compare sensors

So it seems that Canon has made a design decision that 11 stops DR performance for a sensel is sufficient. Just don't go boosting your shadows too far with a Canon DSLR, if you've shot at a low ISO hoping to maximise performance.

Dan.

---

Last edited by dosdan; 12-29-2010 at 04:15 AM.

[enoeske]

We use Canon 1DMII's at work and they are terrible even at low ISO. Super grainy and not all that sharp. They work fine for the size we need (600px for web) but we bring in the 5DMII for model shoots because the 1D can't handle it. I keep meaning to bring in my Pentax K20D with my smc-K 50mm f/1.4 to show em what a real camera can do.

[dylansalt]

We use Canon 1DMII's at work and they are terrible even at low ISO. Super grainy and not all that sharp. They work fine for the size we need (600px for web) but we bring in the 5DMII for model shoots because the 1D can't handle it. I keep meaning to bring in my Pentax K20D with my smc-K 50mm f/1.4 to show em what a real camera can do.

Really????

Tell that to the hundreds of fashion togs across the globe who use the mk111 and many upgrading to the mkiv!

[enoeske]

Maybe the mIII is a lot better than the mII, I certainly hope the mIV is. I am really really unimpressed with the mII.

[jct us101]

I have never had this problem with a Canon DSLR at low ISO?

[Miguel]

So is this post and your proclamations within (" Just don't go boosting your shadows too far with a Canon DSLR, if you've shot at a low ISO hoping to maximise performance") based on any uses of a Canon DSLR in your experience? Which models have you used?

And how exactly are you establishing acceptable noise criteria for evaluating an image from a Canon DSLR? Is it based on a screen image, or a fine print, say at 12x18 inches?

M

[enoeske]

Here's a 100% crop from an image. You can see how mangled the details are at ISO 100 under bright studio strobes. Of course, thats all we have here (mkII) so I can't show any comparisons.

Attached Images

View Picture EXIF

File Date

 

[dosdan]

Consider also this comment from Emil Martinec in his treatise on "Noise, Dynamic Range and Bit Depth in Digital SLRs" (my bolding):

The two-stage amplification employed by Canon gives a bit more information as to what portion of the read noise can be attributed to different circuit components in its high-end cameras. Because of the second stage of amplification, one has separate information on how much of the read noise is due to noise in the main ISO amplification and how much is downstream from it in the secondary amplification and ADC. One sees from the table that the dominant noise at low ISO arises from the main ISO amplifier noise R1, with a smaller contribution R2 from the components downstream from this amplifier and a negligible contribution from the sensor readout noise R0. Lower noise amplifiers would improve low ISO read noise and result in higher dynamic range at low ISO.

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html#read_vs_iso

Dan.

[dosdan]

One of of Roger Clark's main interests is Scientific Astrophotography. Here the read noise at higher ISOs, where the DR naturally becomes more restricted, is of most importance. He measures 8 stops DR for the ID IV at ISO12800. The Nikon D3x hasn't been measured by Clark. The D3x has a high ISO extension which is not enabled in the DxO tests. Extrapolating the ISO6400 curve, the D3x's DR, under the Screen tab at ISO12800, is about 7 stops. Under the Print tab, the difference reduces to about 0.4 stops.

However of more interest in Landscape, Portraiture & General Photography is low-ISO DR performance where the amount of detail hidden is the deep shadows of a typical daylight scene is important.

If Canon improved their electronic read noise (it's been like that for a long while now), their measured, and also I presume their actual, low-ISO DR performance, particularly for their top-end models with their excellent low-noise sensors, would improve significantly.


Dan.

[dosdan]

John Sheehy in the DPR Open Form, has provided an explanation of why most CMOS cameras have not been able to break the 11.5 stops @ ISO100 barrier, except ones using Sony sensors (remove space in URL):

http://forums.dpreview .com/forums/readflat.asp?forum=1018&thread=37308560

He says the typical max. SNR for CMOS sensors at low ISO is 2,800:1. In stops, that's log 2800/log 2 = log 2800 / 0.301 = 11.45 stops.

DxOMark DR figure, on a per pixel basis ("Screen" tab), shows the D7000 has a DR of 13.35 stops at ISO83 (actual) & the K-5 has 13.18 stops at ISO91 (actual). Applying a correction factor to get to true ISO100, these two sensors ISO100 DRs are 13.04 stops (Pentax) & 13.08 stops (Nikon) which, given experimental error, are in damn close agreement. The DR for a midway figure of 13.06 stops is 8,532:1. Since the cameras get higher DRs at actual ISOs that are lower than the manufacturer's ISO100, this indicates that the sensel's full well capacity (saturation) is not yet reached at true ISO100. That's why the D7000 can get an increased DR at ISO83 of 13.35 stops (10431:1). So there's still more in the full well capacity tank and the K-5 manages to go further and reaches 13.61 stops (12,500:1) DR @ ISO80. (indicated) / ISO70 (actual).

Sony's ADC-on-sensor design is a technological breakthrough in the low-ISO area. Pentax, Nikon & Sony are starting to use it. Canon will need to make a major change in their design, rather than a further polishing of its current circuit topology, to match this performance.

Dan.

[Jebus]

here is a nice image taken by a canon user with a 1D mkII at ISO 100....

[palmor]

I always take these types of tests with a grain of salt because in real world final photos you won't see any differences. I've owned a Pentax *ist DS, K10 and K20 and now have a Canon 40d, 7d and 1DMKIII and in the final products there is no difference in low ISO noise.

Sure, you can see some low ISO noise in the 7d if you look at it at 100% viewing but as usual that isn't representative of the quality of the final product.

As always, just IMO

John

[mikemike]

Canons are noisy at low ISO Because thats how PROFESSIONAL photographers like it. Pentax is once again behind the power curve.