Does it make sense? An $1800 Full frame DSLR from Canon

[Daniel Browning]

It's not just pixel size that reduces noise, it's the on-chip (or software) noise reduction algorithms.

I'm aware of when and where noise reduction occurs.

If you took the raw data (without noise reduction being applied) of two chips (with same physical size), the one with larger pixels will appear to be less noisy.

I disagree.

There is always a tradeoff when changing a single parameter.

Yes, there is always a trade off, but noise is not one of them. The trade off for smaller pixels is stronger in-camera processing power, more storage space, slower demosiac and post processsing, etc.

Smaller pixels (higher MP count): bigger enlargements but more noise at
high ISOs.

I think that is a common misconception. For a given sensor size the "noise per detail" (i.e. noise power per spatial frequency) stays the same no matter what the pixel size. The only difference is that smaller pixels allow one to use higher spatial frequencies. This is illustrated in the link I provded above for the 50D/40D.


http://luminous-landscape.com/forum/index.php?showtopic=29801&view=findpost&p= 241562


When you compare them at different spatial frequencies (detail) by using 100% crop, the 50D appears to have more noise. But when you compare them at the same spatial frequency (amount of detail) by resampling both to the same resolution, it becomes clear that their noise is in fact the same.

[ShutterbugJohan]

Daniel Browning said:


"Interesting. I haven't heard that before, but I do know that dead
pixels on higher MP sensors affect the total image much less than the
same number of dead pixels on a lower MP sensor."


Thanks for the responce, Daniel. That White Paper is very informative. What is the purpose of dead pixels? I believe that my Canon 10D has 6.5MP, but is only uses 6.3MP. Thanks.


--Johan

[Daniel Browning]

What is the purpose of dead pixels?

Every sensor has a number of dead pixels and/or hot pixels, where they are full brightness even in the absence of light. They can be caused by impurities in the silicon. Most of them are "mapped out" at the factory. Mapped-out pixels are interpolated from the surrounding pixels. New ones often develop after the camera is shipped, resulting is a surprised customer, until the user learns how to map out the bad pixels themselves, like so:


http://community.the-digital-picture.com/forums/t/673.aspx

[HiFiGuy1]

<p class="MsoNormal"]<span style="font-size: small; font-family: Times New Roman;"]So no downside then!<span style="mso-spacerun: yes"] The cropped lenses from any manufacturer rate along side betamax, hd-dvd, and French cars, just another con to get people to buy and then be forced to buy again!
<div style="CLEAR: both"]</div>

Scotty,


I don't know if I can agree with you on the comparison of EF-S lenses with HD DVD and Betamax. Generally speaking, EF-S lenses are considered to be slightly (or more) inferior when compared with L lenses. In those two examples, the productswere arguable superior to their competition, but lost in the marketplace. French cars, OTOH, I can't speak for, though I'd like to find a really low mileage, clean Renault R5 one day, even if just for a test drive.

[Jon Ruyle]

I disagree. If you have a larger *sensor*, you're going to have lower noise: pixel size is almost irrelevent.

I would have guessed read noise would be larger when you have more pixles (so high pixel density would hurt snr). Are you saying that isn't true? Or that it is true and isn't a significant factor?

[Colin]

He's saying that if you look at signal to noise per pixel, it is true. However, if you look at noise per sensor area,the pixel size is irrelevant. If you look at noise per image area, the larger sensor willl have an advantage regardless of the pixel size.


At least, that's what I retained from a previous related post []


quick example. Half the pixel dimension, 4 small pixels for every 1 big pixel. 1 Big pixel has 4 times the signal to noise as 1 small pixel, because it's got 4 times the signal (4 times as much light). HOWEVER, 4 small pixels, same area, have the same signal to noise ratio as the 1 big pixel, because they get the SAME amount of light, total. If yousum the output of the four small pixels, and average, the signal sums, and the noise cancels out. You lose detail, you increase signal to noise. What more pixels allows you to do is filter more selectively, after the fact, as opposed to filtering mechanically by larger pixel size.

[Daniel Browning]

I would have guessed read noise would be larger when you have more pixles (so high pixel density would hurt snr).

You're in good company. That's the commonly accepted viewpoint among most photographers, web sites, magazines, and just about anywhere you look. But I still think it's incorrect, and it persists due to a fundamental misunderstanding of scale: chiefly, that pixels are compared at 100% crop, where smaller pixels are unfairly examined at higher magnifications and higher spatial frequencies. If all pixel sizes were examined at the same magnification, same print size, same resolution, and same spatial frequency, the bias against small pixels would never have gotten off the ground.

Generally, I think read noise is not correlated to pixel size. For example, the Panasonic LX3 has a read noise of 5.6 electrons per pixel (2.55 ADU), compared to 23.5 electrons in the 5D2, both at base ISO. But that's comparing very different spatial frequencies. After resampling the small LX3 pixels to the much larger 5D2 pixels, read noise per output pixel goes down significantly, because random noise adds in quadrature, so the LX3 is even better. This can be tested by measuring noise on any raw image, then resampling with a good algorithm (e.g. lanczos; not the poorly implemented algorithms in photoshop), and measure the read noise again.

[Daniel Browning]

If yousum the output of the four small pixels, and average, the signal sums, and the noise cancels out. You lose detail, you increase signal to noise. What more pixels allows you to do is filter more selectively, after the fact, as opposed to filtering mechanically by larger pixel size.

Excellent summarization, Colin.


For an example, take a comparison between pixel sizes of 4 microns and 2 microns:

• If the 2um pixel has read noise that is twice as bad, then the final image has the same read noise.
• If the 2um pixel has read noise that is the same, then the final image has much less read noise.
• If the 2um pixel has more than twice as much read noise, then the final image has more read noise.

So smaller pixels can be noisier per pixel (by an amount equal to the square root of the difference in size) to achieve the same amount of noise as larger pixels. My position is that, all other things being equal, smaller pixels do tend to be just about that much noisier, and so the final image is just about the same.

[George Slusher]

The "cropped" (EF-S) lenses are necessary to give the "cropped" bodies similar performance to full-frame bodies at wide angles. Seen many 10-22mm zooms for full-frame bodies? The 16-35mm gives the same field-of-view on a full-frame body. The EF-S lenses are also probably less expensive to make for similar performance (smaller sensor size means that it's easier to control distortion, vignetting, etc.). Compare the EF-S 17-55mm f/2.8 IS ($1030 at B&amp;H) vs the 16-35mm f/2.8L ($1450). (Read Bryan's review to find out that the 17-55mm lens "matches or exceeds the optical performance of my L-series zooms in this similar focal length range," plus it has IS.)


The "cropped" bodies are not going away, so it's not like Betamax or HD-DVD. I have no data, but I would expect that the 1.6x FOV cameras from Canon outsell the full-frame cameras by at least 10-to-1, probably a lot more.

[Sinh Nhut Nguyen]

Guys, I think we need to give this up and go shoot some photos yeah?[]