Why *less* MP is better

[Lee Jay]

Electronics collect the photons, convert them to electric signals
and low-noise ampilify the signals to create the image. The process
is light-to-electric signal conversion and amplification. As you
work with smaller size microelectronics the more efficient you can
make this process per surface area to get a higher signal-to-noise
output.

Okay, let's say the electronics are 100% perfect. Will there be any noise in the image?

--
Lee Jay
(see profile for equipment)

 

[Bob Blount]

I was fortunate to get a national mag cover with a 3.2MP, all auto, Nikon 3100. It is not the camera but how it is used. A 5D in the wrong hands will not assure quality pics.
--
Bob,
Great Pro 1 + 420EX
Wife's Nikon 3100

 

[Bob Blount]

http://www.kenrockwell.com/tech/notcamera.htm
--
Bob,
Great Pro1 + 420EX
Wife's Nikon 3100

 

[Lee Jay]

http://www.kenrockwell.com/tech/notcamera.htm

There is definitely some truth to this "it's not the camera, it's the photographer" thing. However, the usage is usually wrong. What is usually implied is that a great photographer will get just as good of shots with lousy equipment as with great equipment. This is wrong.

I think it's usually true that a great photographer with lousy equipment will out-perform a lousy photographer with great equipment. However, a great photographer will perform better with great equipment than with lousy equipment.

Also, there are shots that simply cannot be captured without certain equipment. Shallow DOF, extreme wide or long, low-light, bursts, and fast closing speeds (requiring AI servo) come to mind. Here are a few of those:

50mm, f1.4 on a 20D:



15mm fisheye, f2.8, ISO 3200 on a 5D:



95mm, 1/20th handheld with IS, f2.8, ISO 800 on the 20D:



150mph closing speed on a 20D, 70-200 2.8L IS with stacked 1.4x TCs:



--
Lee Jay
(see profile for equipment)

 

[Bob Blount]

Findger wrote, "What is usually implied is that a great photographer will get just as good of shots with lousy equipment as with great equipment. This is wrong."

That is not the point of the author but he is usually misquoted to dismiss his true point of view. I hope others will read it for themselves rather than count on others to discredit the author.

--
Bob,
Great Pro1 + 420EX
Wife's Nikon 3100

 

[mamallama]

Electronics collect the photons, convert them to electric signals
and low-noise ampilify the signals to create the image. The process
is light-to-electric signal conversion and amplification. As you
work with smaller size microelectronics the more efficient you can
make this process per surface area to get a higher signal-to-noise
output.

Okay, let's say the electronics are 100% perfect. Will there be
any noise in the image?

Yes, here's why. When the signal is in the analog form there will always be noise in the signal. But that noise will only be insignificant after digitization if it is below the quantization level in the analog-to-digital (A/D) process. That's generally a big order, but as the analog signal-to-noins ratio (SNR) is improved with better/smaller microelectronics, it can be much improved.

Digital electronics are perfect, only ones and zeros to amplify and transfer. But the digital elextronics inherit noisy bits if the SNR of the analog signal is high relative to the quantization level in the A/D process.

--mamallama

 

[orpheo]

Very interesting post and discussion here. I basicly agree with what LeeJay says in his original post.

And I've read an article that states, that about 6MP resolution would be enough for ANY size of (big) picture, given the optimal distance to view it from (the human eye can only discern so much detail from a certain distance and a bigger picture is usually viewed from a bigger distance to comprehend the content of the picture - not to count pixels). They're talking of "real" resolution, which is probably best represented by lph-measurements and not the pixel-count of the image file. So maybe the resulting lph of a sensor-lens combination at a certain ISO-Level would be the truer value for marketing than the pixel-count.

Now these are all measurable facts dependent on the laws of physics. Perceived image quality is not a measurable fact. Some refer to the joy of grandparents, the sound quality of loudspeakers or dream about the specs they would like a camera to have. Dream on, it will propably come true some day... (And perceived quality of music-reproduction is one of the least objective themes I know of.)

Meanwhile, I think, prosumers would be better off with cameras that produce usable physical picture-quality in terms of real resolution (including the aspect of noise) and dynamic range in common situations. Which is not the always-perfect-light or the I need 300dpi for magazine-prints from a compact-camera situation.

So, the question that puzzles me the most is: why does marketing so often choose the wrong aspects to promote?

 

[mamallama]

Lee Jay:

If you think I'm making all this up, here's an article that describes a new CMOS imaging chip under development that does indeed take advantage of, among other things, smaller size transistors on the chip to leave more chip real estate for the photo diodes. The new chip produces better images, uses less power and has a dynamic range 100 times better than current sensors! That's big.

http://www.dpreview.com/news/0512/05121201new_chips.asp

There's still some improvements in electronics left. We haven't gone as far as we can go yet before we have to worry about the end of the universe and changing the laws of physics.

--mamallama

Not the end of the universe nor any change in the laws of physics.
Just that solid state technology and microelectronics developments
are not yet at the molecular level but approaching fast.

The electronics aren't the problem. The lack of available photons
is a problem.

--
Lee Jay
(see profile for equipment)

 

[Lee Jay]

Okay, let's say the electronics are 100% perfect. Will there be
any noise in the image?

Yes, here's why.

[snip wrong explaination]

You gave the right answer for the wrong reasons. Read this to understand shot or photon counting noise:

http://en.wikipedia.org/wiki/Photon_noise

--
Lee Jay
(see profile for equipment)

 

[Tom O'Neil]

Once again we are hostage to the "vast majority." My portable snapshot camera is an S45. I will never "upgrade" to an S80 because Canon decided the vast majority didn't know what RAW mode was and didn't know how to use it.

yes, its true 1MP is going to be capable of printing 4x6, though
most likely it will still be slightly on the grainy side. 2mp is
more than enough for 4x6 though, so why have compacts at 5mp and
7mp?
the reason is the crop and edit.

The problem with this is that the vast majority of purchasers of
these smaller, less-expensive cameras don't ever crop or edit. I
see people standing at the Costco stations for seemingly hours
trying to figure out how to get a 4x6 glossy from their 5MP shots
never getting past the phone number screen. Don't confuse dpreview
forum members with the general public! Purchasers of the
"prosumer" cameras are different and the cameras themselves are
different too. The 3x zooms are limiting for framing purposes but
the 28-200 on the Pro1 is much less so.

i don't know how many times you
can frame a shot perfectly with only 3x zoom on a camera that has a
wide-angle 28mm lens already. your 2mp jepg won't be able to be
cropped at all without significant loss of image quality. at the
very least, to print some 4x6 cropped, you'll need 3.2MP.

with that said, many people are also going to want larger prints.
you made it sound like we all print on our deskjets and only tiny
little pictures. in the past year, i've had 5 16x20's printed and
15 8x10's printed (imagestation). if you're really enjoying and
into photography, there's a good chance you'll want more than a
stack of 4x6's as a result of your work. even beginners like to see
the fruit of their labor nice and large.

I did mention up to 13x19 in my post for more advanced users
targeted by the "prosumer" cameras. 16x20 isn't much bigger for a
4:3 camera (12% more linear resolution needed for the same ppi).

--
Lee Jay
(see profile for equipment)

 

[mamallama]

What's wrong? I said that there will always be noise in the signal in the analog forn. That noise will be made up of SHOT and THERMAL noise. But it's not the noise alone that is significant, it's the noise in relation to the signal, the signal-to-noise ratio (SNR). That SNR will determine how sensitive (Max ISO) you can use the sensor at. Smaller microelectronics are allowing larger photo diodes on the same small chips to get better SNRs as is now being developed at the U. of Rochester.

What's wrong about that? Or don't you understand the process? Read about that Rochester chip I referred you to and you may get a better understanding of what I said.

http://www.dpreview.com/news/0512/05121201new_chips.asp

--mamallama

Okay, let's say the electronics are 100% perfect. Will there be
any noise in the image?

Yes, here's why.

[snip wrong explaination]

You gave the right answer for the wrong reasons. Read this to
understand shot or photon counting noise:

http://en.wikipedia.org/wiki/Photon_noise

--
Lee Jay
(see profile for equipment)

 

[Lee Jay]

What's wrong? I said that there will always be noise in the signal
in the analog forn. That noise will be made up of SHOT and THERMAL
noise. But it's not the noise alone that is significant, it's the
noise in relation to the signal, the signal-to-noise ratio (SNR).
That SNR will determine how sensitive (Max ISO) you can use the
sensor at. Smaller microelectronics are allowing larger photo
diodes on the same small chips to get better SNRs as is now being
developed at the U. of Rochester.

Smaller electronics will not allow better SNR. It will allow slightly better dynamic range. The reason you don't get better SNR is that the micro lenses already gather the light from the area over the non-sensitive area of the chip and focus it onto the sensitive area. Therefore, the micro-lens design (and size, obviously) is the driver for how many photons are collected, not the area of the sensing element.

What's wrong about that? Or don't you understand the process? Read
about that Rochester chip I referred you to and you may get a
better understanding of what I said.

http://www.dpreview.com/news/0512/05121201new_chips.asp

That approach will have no real effect on noise at all. There are several cleaver approaches for getting better dynamic range including my personal favorite (at this time) of re-using the cell once it's full. None of these are yet available but, if they do become available, then dynamic range can be effectively infinite changing the story back to just a matter of noise. In that case, the number of pixels is largely irrelevant. In fact, you might be able to get better noise performance with more pixels of a smaller size than with fewing larger pixels. Like I said above, the dynamic range is the problem.

--
Lee Jay
(see profile for equipment)

 

[technicsplayer.]

So, the question that puzzles me the most is: why does marketing so
often choose the wrong aspects to promote?

because marketing people are not the designers and generally in my experience are a lot more stupid.

 

[Ron Schroeder]

Hi Lee,

I mostly agree with you but I feel that 2MP is the minimum for a 4x6. The reason that I say this is that even with film, the one hour photo labs print a 4x6 at about 2MP. I sometimes joke that the only way I can tell the differance between a shot from a Canon A60 (2.1MP) and a 35mm point and shoot is the print from the digital file doesn't have scratches or dust specs.

When these photoprocessors print film, they still scan the negative to creat a digital file. The file is massaged for color balance, brightness etc and then is scanned onto photographic paper and run thru the "soup". With a 4x6, that scanned image is about 2MP, no matter how fine grained the negative was.

Make it 3MP and even with some cropping, a 4x6 from a digital camera will be as sharp as film.

What I would really like would be a P&S with either a 2/3" 4MP sensor or better yet, a 4/3 5MP sensor and the latest electronics.
--
Ron Schroeder

 

[mamallama]

What's wrong? I said that there will always be noise in the signal
in the analog forn. That noise will be made up of SHOT and THERMAL
noise. But it's not the noise alone that is significant, it's the
noise in relation to the signal, the signal-to-noise ratio (SNR).
That SNR will determine how sensitive (Max ISO) you can use the
sensor at. Smaller microelectronics are allowing larger photo
diodes on the same small chips to get better SNRs as is now being
developed at the U. of Rochester.

Smaller electronics will not allow better SNR. It will allow
slightly better dynamic range. The reason you don't get better SNR
is that the micro lenses already gather the light from the area
over the non-sensitive area of the chip and focus it onto the
sensitive area. Therefore, the micro-lens design (and size,
obviously) is the driver for how many photons are collected, not
the area of the sensing element.

What's wrong about that? Or don't you understand the process? Read
about that Rochester chip I referred you to and you may get a
better understanding of what I said.

http://www.dpreview.com/news/0512/05121201new_chips.asp

That approach will have no real effect on noise at all. There are
several cleaver approaches for getting better dynamic range
including my personal favorite (at this time) of re-using the cell
once it's full. None of these are yet available but, if they do
become available, then dynamic range can be effectively infinite
changing the story back to just a matter of noise. In that case,
the number of pixels is largely irrelevant. In fact, you might be
able to get better noise performance with more pixels of a smaller
size than with fewing larger pixels. Like I said above, the
dynamic range is the problem.

--

Did you read this, or don't you understand? It's all about SNR, not just noise alone. Also note that 100 times improvement in dynamic range that you said is a problem. From the U. of Rochester chip description:

"Traditional image sensors use an array of light-sensitive diodes to detect incoming light, and transistors located at each photodiode to amplify and transmit the signal to an analog-to-digital converter located outside of the photodiode array. Other designs can convert the signal to digital at the pixel site, but require high precision transistors, which take up considerable chip space at each pixel and reduce the amount of surface area on the chip devoted to receiving light. The new design not only uses smaller transistors at each pixel, and thus can allow more light to be detected, but the transistors can be scaled down in size without diminishing the sensor performance as advances in semiconductor fabrication technologies allow the size of transistors to shrink. This means that much denser, higher-resolution chips can be developed without the prohibitive problems of the existing sensor designs. When transistors are reduced in size, they also become faster, allowing incoming light to be sampled more frequently and accurately."

"The first technology being developed integrates an oversampling "sigma-delta" analog-to-digital converter at each pixel location in a CMOS sensor. "CMOS" is a common semiconductor fabrication process used in most chips manufactured today. Previous attempts to do this on-pixel conversion have required far too many transistors, leaving too little area to collect light. The new designs use as few as three transistors per pixel, reserving nearly half of the pixel area for light collection. First tests on the chip show that at video rates of 30 frames per second it uses just 0.88 nanowatts per pixel--50 times less than the industry's previous best. It also trounces conventional chips in dynamic range, which is the difference between the dimmest and brightest light it can record. Existing CMOS sensors can record light 1,000 times brighter than their dimmest detectable light, a dynamic range of 1:1,000, while the Rochester technology already demonstrates a dynamic range of 1:100,000."

--mamallama

 

[Lee Jay]

I did read it and I think I understand it. I'm saying that it will not effect low-light sensitivity or high-ISO noise. It has the capability to effect bright-light dynamic range because of its ability to preserve highlights.

The only way to significantly improve the high-ISO performance of today's sensors without making them bigger, is to improve quantum efficiency. That's because they are already limited by shot noise. Video people know this and that's why there are so many 3-CCD camcorders. Removing the Bayer mask improves overall QE by a factor of roughly 2 giving 3-CCD camcorders roughly 1 extra stop of low-light performance.

No amount of on-chip electronics tricks will reduce shot noise unless they improve QE. The approach you linked to will not improve QE.

--
Lee Jay
(see profile for equipment)

 

[Lee Jay]

Hi Lee,

I mostly agree with you but I feel that 2MP is the minimum for a
4x6.

Notice that I used terms like "pretty decent", "very good" and "pretty good" and avoided terms like "perfect".

--
Lee Jay
(see profile for equipment)

 

[mamallama]

I did read it and I think I understand it. I'm saying that it will
not effect low-light sensitivity or high-ISO noise. It has the
capability to effect bright-light dynamic range because of its
ability to preserve highlights.

With larger photosites your sensitivity goes up.

The only way to significantly improve the high-ISO performance of
today's sensors without making them bigger, is to improve quantum
efficiency. That's because they are already limited by shot noise.
Video people know this and that's why there are so many 3-CCD
camcorders. Removing the Bayer mask improves overall QE by a
factor of roughly 2 giving 3-CCD camcorders roughly 1 extra stop of
low-light performance.

"quantum efficiency"? What's that a buzz word for? BTW we are not talking about TODAYS SENSORS. These are new TOMORROW SENSORS. They are yet to come.

No amount of on-chip electronics tricks will reduce shot noise
unless they improve QE. The approach you linked to will not
improve QE.

You're still missing the essence of the design. No need to reduce shot or any noise, you're increasing the signal by having larger photosites on the same size chip because the smaller electronics take up less space. It's the SNR, not the amount of noise. Period.

--mamallama

 

[Lee Jay]

With larger photosites your sensitivity goes up.

Ummm...no. With larger capture area , sensitivity goes up. The capture area is that covered by the microlenses.

"quantum efficiency"? What's that a buzz word for?

It's the number of electrons produced divided by the number of photons entering. Ideally, you'd like each photon to produce 1 electron. Unfortunately, the Bayer mask reduces the light that hits the micro lenses, the micro lenses aren't perfect, and the sensor itself is not 100% efficient for a host of reason. So today's sensors are in the 20% range on QE. The best monochromatic sensors are above 80%.

No amount of on-chip electronics tricks will reduce shot noise
unless they improve QE. The approach you linked to will not
improve QE.

You're still missing the essence of the design. No need to reduce
shot or any noise, you're increasing the signal by having larger
photosites on the same size chip because the smaller electronics
take up less space. It's the SNR, not the amount of noise. Period.

Except that you aren't getting more signal because the micro lenses are the limit, not the sensing element. A bigger sensing element gives you more well capacity which helps a bit with dynamic range in bright environments, but it doesn't help at all with high ISO sensitivity because well capacity isn't the limit.

You should look at this:
http://web.canon.jp/Imaging/cmos/technology-e/light_gathering.html

--
Lee Jay
(see profile for equipment)

 

[Lee Jay]

http://en.wikipedia.org/wiki/Quantum_efficiency

--
Lee Jay
(see profile for equipment)