Rumors > EOS Bodies

9% larger pixel size on 1D X - how does that translate IQ-wise?

(1/8) > >>

Redreflex:
Here's a comparison of relative pixel sizes. Quote from Chuck Westfall (Canon USA Pro Engineering division) via dpreview: (http://www.dpreview.com/articles/5149972341/canon-eos-1d-x-overview)

When asked to for the biggest improvement in the new camera, Westfall stresses that every aspect of the camera has been re-assessed but finally concludes: 'If you had to highlight just one thing, I'd say the sensor. It's a new level for us in terms of image quality.'

'There's a couple of things that we consider when we think about IQ: number one on this sensor is noise. It's clear the noise level is better than in the 1D Mk IV or the 1DS III. The pixel size is larger than in the 1DS III or 5D Mark II (6.95 microns, versus 6.4) and the difference is even more striking compared to the 5.7 micron pixels in the 1D Mark IV. That helps us in terms of light capturing ability and increases the signal to noise ratio. In turn, that does nothing but help the dynamic range of the camera.'

The pixel size is 8.5% larger in the 1D X compared to 1Ds3 or 5Dm2.

Everything else being equal, how does that translate into real-world image quality improvement? (I accept there'll be other factors to consider other than pixel size)

Gothmoth:

--- Quote from: Redreflex on October 18, 2011, 05:38:38 AM ---The pixel size is 8.5% larger in the 1D X compared to 1Ds3 or 5Dm2.
--- End quote ---

the diameter is 8.5% larger.

Edwin Herdman:
Chuck Westfall demonstrates he isn't an engineer; nothing else.  He's a good guy but don't forget he's a PR guy and these numbers might as well have been pulled out of a hat as far as his description of them here.

I accept as true that at some point smaller-size pixels do run into IQ problems.  The actual point where you'd draw the line for IQ is not guaranteed to be the point where they are drawing the line, because they have to produce these sensors in bulk and so a smaller pixel pitch could translate into higher reject rates for chips.

I think Gothmoth is right here; this is discussing (as usual) pixel diameter.  9% would be less than a generational change in technology would improve ISO performance, but it fares better than 9% in comparison with the Mark IV (less by far in comparison with the 1Ds Mark III though).  I'm not positive my numbers are right, but squaring both micron sizes (1D X versus 1D Mark IV) gives me an end ratio of (1D X micron size / 1D Mark IV size) = 1.486... times the sensor size.  You'll note that's a little bit larger than the 1.3x crop factor would allow for (again, I might be making an error here; though it doesn't make sense to normalize one pixel size to the other).  It goes to show that, once again, sheer sensor size trumps other considerations, but otherwise this isn't a straight comparison - for a 1.3x sensor camera to compete on a per-pixel basis (remembering that noise or any other picture attribute is meaningless on a single pixel basis) with the 1D X it would have to have a low megapixel count, by any modern standard, which is why I wondered for a moment if it wouldn't make sense to normalize the Mark IV pixels to their size on a full frame sensor (this would be a silly comparison, of course, since they'd be even slightly larger than the 1D X pixels).

My feeling is that Canon is joining Nikon (actually, they did some time ago as a German rep for Canon described a relationship between noise and pixel density which sounded a lot like the typical Nikon statement) in downplaying the importance of more megapixels, but the real calculations on the tradeoffs between ISO and production being done are likely trade secrets and not revealed.  Furthermore, many shooters simply are not going to buy this public argument because they require as much resolution as possible.  It will be interesting to observe the silence when a 5D Mark II replacement comes out that is also full frame but features a smaller pixel pitch.

The interesting question here, however, is one I would think holds true even if you believe (and I admit it's quite possible) that they actually aren't obfuscating the truth with marketing, and that there is this relationship (and they've only simplified it a bit for the masses) - what are the assumptions about the market use of cameras that leads them to downplay megapixel counts on one camera and not another?  The shooting rate of this new 1D X is limited by more than just the number of pixels on the sensor, such as the CPU (in this case, interestingly, we see that the new DIGIC 5 does indeed come in various "flavors," this one being the "5+" but still used in a pair for the new camera, so perhaps a sole 5+ will be the CPU for more mainstream DSLRs, and the "DIGIC 5" may be for compacts, if I have not misread this) and the actual hardware of the SLR mechanisms - in combination, a fast mirror return rate with a high megapixel count sensor would send much more data to the sensor than seen here.  If they're using two DIGIC 5+ CPUs, we don't have proof that these CPUs are being taxed to the limit, but it doesn't seem likely that just one would suffice.

9% larger pixels seems less significant to me (and I think the numbers back this up) than just the typical generational change in the sensor production technology and other patented elements that are esoteric and do not follow this easy (and I think misleading) apparent mathematical relationship - like improving microlenses, or most importantly improvements in the photon to signal capture chain (including the ADC).

Ivar:
The only thing needed to be be done for a proper comparison between the technologies (NB! not sensor sizes) is to compare the output for exactly the same sensor area.

Thus, take a high ISO shot (highest native for the 1D mk4 as to stay at the hardware level) with both, with the 1D Mk4 & 1Dx, crop the latter one to the 1.3x (the 1Dx has then 13.85MP left out of 18MP) and downsize the bigger 1D Mk4 file to 13.85MP.  This way, it is possible to compare the technologies rather than the cameras.

The results - the 1Dx might be better per area, but it is not as good as marketing tries to show, it is the sensor size what adds most to the equation, not the individual size of the sensel. 

P.S obviously the real test can be only done when the new X is available.

NotABunny:

--- Quote from: Redreflex on October 18, 2011, 05:38:38 AM ---Everything else being equal, how does that translate into real-world image quality improvement?
--- End quote ---

Umm, but nothing else is equal. They are launching a new platform, DIGIC 5: processor and sensor technology.

It's far more interesting to look at the maximum native ISO: 51200. They are confident that this is the maximum native ISO which delivers acceptable quality when a shot must be taken. The maximum native ISO of the 5D2 is 6400, so 3 stops more is mind-bending.

Navigation

[0] Message Index

[#] Next page