more pixels are better!

or are they?

Could someone try to give a clear but comprehensive explanation of why more Mp doesn't improve IQ?

I understand that more pixels means more noisy pixels, and i've seen that in my own cameras.

But should twice as many pixels give twice the dynamic range, simply because you could sample two pixels for a value with twice the range of a single pixel?

Isn't it a question of whether NR is losing detail that wouldn't be there in the first place with less pixels?

One day I penned a piece about that. It is here:
http://www.dpreview.com/articles/compactcamerahighiso/

I am not going to update it, because it is not absolete. The other reason is, that nobody is reading it anyway,
How did you get your sense of the news?
(-)

Image detail is affected by lens aberrations...
http://www.u.arizona.edu/~nsophia/report.htm

...and the size of the sensor.
http://www.cambridgeincolour.com/tutorials/digital-camera-sensor-size.htm
(and many other references)

Technically, Dynamic Range is the differences between the smallest and largest recognizable signal that can be recorded by a pixel. Thus larger pixels have greater dynamic range than smaller pixels. However, this strict definition doesn't take into account the combined effect of pixels. I personally remain uncertain as to the effect of pixel size and density on dynamic range. I've seen examples that indicate larger pixels give more dynamic range for a given sensor size, and I've seen examples that indicate larger sensors give more dynamic range for a given pixel size.

Bottom line...bigger is better. You want better detail, get bigger lenses and bigger sensors.

Notice my reference to image detail rather than image quality. A quality image is something else entirely!

Dear I Forget,

There is so much disinformation about this, it is not surprising people get confused. Unfortunately, one of the main disinformers is DPR. So, here goes:

All else being equal, more MP does indeed improve IQ, or at least, give the potential for improved IQ. Other factors also affect IQ, in different ways. In the end, in a digital camera, IQ depends of just four things:

i) The ability of the lens to direct photons reflected by the lens to the right place in the image plane.

ii) The ability of the sensor (plus it's optical low pass filter) to accurately record the number and location of those photons.
iii) The number of photons that go up to make an image.

iv) The ability of the sensor to record accurately the number of photons of different wavelengths such that an image can be reconstructed that stimulates the colour vision of the eye in the same way as the original scene.

Number (iv) you will notice is rather complex, and seems to be becoming a more discussed part of the IQ equation. Nonetheless, I'm going to ignore it here, since it is not so dependent on pixel count.

Looking now at number (ii), the more pixels a sensor has, the more accurately it can record the location of the photons landing on it. By itself, changing the number of pixels does not affect a sensors ability to record the total number of photons, so to that extent, more pixels increases IQ, absolutely.

The other issue in the IQ equation is the size of the sensor. At first sight, a larger sensor captures more photons, so should result in higher IQ. However, it's not that simple. If you aim to take exactly equivalent photos (that is with the same field of view, depth of field and shutter speed) with two cameras with different sensor sizes, it turns out that the number of photons going into the image is exactly the same, so from that point of view, all sensor sizes produce the same IQ. However, while this is theoretically true, generally the bigger the sensor you have the higher IQ you can produce. There are two reasons for this. Firstly, people don't always shoot equivalent images. Professional and enthusiast cameras tend to have better focussing systems, which mean people shoot with less depth of field which results in larger apertures and therefore more photons in the image. Secondly the resolution of the lens is exercised less by larger sensors, and in any case, lenses for professional/enthusiast systems tend to produce more resolution in realtion to sensor size than the smaller sensored 'point and shoots'.

Issues such as diffraction and per-pixel noise, which are often quoted as a determinant of IQ, are frankly a complete red herring. they only affect things if you compare cameras in non-equivalent situations, like different output image sizes or different dept of field. Keep things equivalent, and all else being equal the more pixels the better the IQ. Of course, all things aren't always equal, some sensors are more efficient at converting photons to signal than others. If a big, high pixel count sensor is very inefficient, it may be beaten by a more efficient sensor which is smaller, or has less pixels.

I know that a whole load of people are going to say what i have written is rubbish. They may say that, but none will be able to post a concrete example illustrating that it is.

Iforgetwhat8was4 wrote:

or are they?

Could someone try to give a clear but comprehensive explanation of
why more Mp doesn't improve IQ?

I understand that more pixels means more noisy pixels, and i've seen
that in my own cameras.

But should twice as many pixels give twice the dynamic range, simply
because you could sample two pixels for a value with twice the range
of a single pixel?

Isn't it a question of whether NR is losing detail that wouldn't be
there in the first place with less pixels?

boels069 wrote:

compensates for noise:
http://www.dxomark.com/index.php/eng/Insights/More-pixels-offsets-noise !

"If you keep the same optics, the same sensor size (and therefore the same field of view), what will be the result? Basically, the same amount of light captured by the optics will be shared by more pixels, and since each individual pixel will receive less light, the signal-to-noise ratio (SNR) will be lower. In short, you will end up taking photos with more noise."

This article is why I don't entirely accept DXO Labs as trusted source just yet. This statement is wrong. The pixels may have more noise...but the photos, viewed as a whole, will have the same noise (or less, assuming that new, same-sized sensor has been improved.)

Also, the technology in the 1DsMIII sensor is a pretty big leap from the 350D. No matter how they try to equalize it, I can't take a comparison like that seriously.

“Suppose now that the images have exactly the same field of view (by changing the lens focal length) and are printed on 30x20cm paper with the same 300dpi printer. Because of the very high resolution of the 1Ds Mark III, the printer will downsample the image and decrease the noise, giving a clear advantage (about 3dB) to the 1Ds Mark III.“

But if the sensors were the same size, then the high MP sensor would have had more noise to begin with. But they're comparing the same image taken with two different sized sensors, which always favors the larger sensor!!

Comparision of RAW images show that the G10, having nearly 50% more pixel than the LX3, has pretty much the same noise performance as the LX3. Obviously there's a problem with DxO's analysis.

Thanks all, for your input.

Let me take another shot at understanding...

For a given sensor size and given chunk of light (more or less base ISO), the camera can fill its pixels with enough photons that it can statistically supress noise. This allows a 12mp compact, in good light, to produce a file that can be enlarged beyond a 6mp DSLR without showing pixelization. (Other limitations of the small format might separate the image quality, however.) This is the easiest way for a compact to rival some of a DSLR's capacities, so the designers are motivated to do so.

Now, if you add more pixels, you cannot capture more information. The noise increases with the number of pixels, and you end up dividing up noise and signal without increasing the latter. The same thing happens when you increase ISO (reduce the chunk of light - the signal). So at higher ISO, the sensor is not optimized to that level of light and binning isn't perfect, so you lose more than you would with fewer Mp to begin with. (???)

There should be a physical and practical limit to the number of useful pixels for a given sensor size and EV, correct?

Am I warm?-)

Iforgetwhat8was4 wrote:

Let me take another shot at understanding...

For a given sensor size and given chunk of light (more or less base
ISO), the camera can fill its pixels with enough photons that it can
statistically supress noise. This allows a 12mp compact, in good
light, to produce a file that can be enlarged beyond a 6mp DSLR
without showing pixelization. (Other limitations of the small format
might separate the image quality, however.) This is the easiest way
for a compact to rival some of a DSLR's capacities, so the designers
are motivated to do so.

No, sorry...this is a strike. Given similar pixel technology, noise depends entirely on sensor size. Given the same framing, the small sensor image will have more noise than the large sensor image. It is possible that the image from the 12MP compact will have more detail. That's certainly a possibility in good light. But even with more detail, it will have more noise. Even if the detail is reduced to match the 6MP sensor, it will have more noise.

Smaller sensor = more noise. There's no way around it. This, along with some other reasons, is why compacts with small sensor can't rival DSLR images (and they really don't...really.)

Now, if you add more pixels, you cannot capture more information.
The noise increases with the number of pixels, and you end up
dividing up noise and signal without increasing the latter. The same
thing happens when you increase ISO (reduce the chunk of light - the
signal). So at higher ISO, the sensor is not optimized to that level
of light and binning isn't perfect, so you lose more than you would
with fewer Mp to begin with. (???)

Hmmm...not how I would put it. At higher ISO you take in less light. Less light causes the noise to be statistically greater. Since the image is darker, the signal/noise combo is amplified to get the correct exposure, but that's not making the noise any worse...the noise was worse in the capture. In any case. Binning and such won't help...if it did all large sensors would be doing it...but they don't...because it doesn't help.

Am I warm?-)

Certainly at least 98.6 F

Graystar wrote:

Iforgetwhat8was4 wrote:

Let me take another shot at understanding...

For a given sensor size and given chunk of light (more or less base
ISO), the camera can fill its pixels with enough photons that it can
statistically supress noise. This allows a 12mp compact, in good
light, to produce a file that can be enlarged beyond a 6mp DSLR
without showing pixelization. (Other limitations of the small format
might separate the image quality, however.) This is the easiest way
for a compact to rival some of a DSLR's capacities, so the designers
are motivated to do so.

No, sorry...this is a strike. Given similar pixel technology, noise
depends entirely on sensor size. Given the same framing, the small
sensor image will have more noise than the large sensor image. It is
possible that the image from the 12MP compact will have more detail.
That's certainly a possibility in good light. But even with more
detail, it will have more noise. Even if the detail is reduced to
match the 6MP sensor, it will have more noise.

Smaller sensor = more noise. There's no way around it. This, along
with some other reasons, is why compacts with small sensor can't
rival DSLR images (and they really don't...really.)

I think we may be saying the same thing. A smaller sensor with more noise but twice the "resolution" (pixel count) can enlarge twice the size of the larger sensor, provided the absolute noise level is small enough. I'm guessing that in good light, this is possible for a P&S, and this is what's driving the designs. (?)

John Carson wrote:

The new Fuji sensor design relies on pixel binning in low light. The
basic objection to pixel binning in general is that same-coloured
pixels aren't adjacent. The new Fuji design changes that so that
same-coloured pixels are adjacent.

“It is well known that increasing the pixel count on a sensor actually makes it more difficult to achieve high sensitivity and wide dynamic range.”

OMG don't give John Sheehy the impression that you support words like that...you'll NEVER hear the end of it!

But seriously, I think you missed this part...

“Even though the sensor has been designed for ‘Dual Capture’ for Wide Dynamic Range and ‘Pixel Fusion’ for Low Noise, it actually performs as well as previous 12-megapixel Super CCD sensors due to the new filter and photodiode design.”

So it seems that this new wonderful sensor only performs AS well as previous 12MP sensors...it's just learned a few new tricks. Current cameras don't bin, so this could be an advantage in terms of feature set. But I think it questionable as to how advantageous binning is to the kind of photography we generally discuss. This may matter to celestial photographers...don't know how much it will matter to others.

Iforgetwhat8was4 wrote:

I think we may be saying the same thing. A smaller sensor with more
noise but twice the "resolution" (pixel count) can enlarge twice the
size of the larger sensor, provided the absolute noise level is small
enough. I'm guessing that in good light, this is possible for a P&S,
and this is what's driving the designs. (?)

Hmmm....don't know about that. I guess it depends on the size difference and type of camera.

For example, consider the Canon G10 and the Canon 50D. By your thinking, the G10's 14.7 MP image, taken in good light and at a low ISO, should be able to be enlarged to practically the same level as the 50D's 15.1 MP image. After all, they're less than half a megapixel apart. However you'd be wrong. The 50Ds image is much better. MUCH better. It can be enlarged further than the G10's image. It's not just that it has less noise. The image itself is more detailed. The combination of larger sensor and bigger lens produces superior detail that the G10 can't match. Throw in the G10's additional noise, and you really are comparing apples to oranges.

Graystar wrote:

Iforgetwhat8was4 wrote:

I think we may be saying the same thing. A smaller sensor with more
noise but twice the "resolution" (pixel count) can enlarge twice the
size of the larger sensor, provided the absolute noise level is small
enough. I'm guessing that in good light, this is possible for a P&S,
and this is what's driving the designs. (?)

The 50Ds image is muchbetter. MUCH better. It can be enlarged further than
the G10'simage. It's not just that it has less noise. The image itself is
more detailed. The combination of larger sensor and bigger lens
produces superior detail that the G10 can't match. Throw in the
G10's additional noise, and you really are comparing apples to
oranges.

Iforgetwhat8was4 is correct (and has been all through). Of course the lens makes a huge difference, but you can't now introduce the lens as a factor, having previously said it was all down to the sensor size. Plus it does not help using undefined terms like "superior detail".

Graystar was also wrong earlier:

"Comparision of RAW images show that the G10, having nearly 50% more pixel than the LX3, has pretty much the same noise performance as the LX3. Obviously there's a problem with DxO's analysis."

On DxO's analysis, at 18% reflectance the G10 has SNR of about 27dB at ISO100 and 18dB at ISO800, but at 1% reflectance it has SNR of about 13dB at ISO100 and 0dB (!) at ISO800. At 18% reflectance the LX3 has SNR of about 30dB at ISO100 and 21dB at ISO800, but at 1% reflectance the LX3's SNR is 18dB at ISO100 and 6dB at ISO800. That is, even in good light the LX3 is one stop better (not "pretty much the same") and in poor light there is no comparison, whcih is exactly what you would expect. You don't pay the full penalty of small pixels until the light goes, so you have to look at the full SNR graphs, not just the figures at 18%.
--
2 November 1975.

'... Ma come io possiedo la storia,
essa mi possiede; ne sono illuminato:
ma a che serve la luce?'

Les Olson wrote:

Iforgetwhat8was4 is correct (and has been all through). Of course
the lens makes a huge difference, but you can't now introduce the
lens as a factor, having previously said it was all down to the
sensor size. Plus it does not help using undefined terms like
"superior detail".

Please show me where I say that IMAGE QUALITY is all about sensor size and nothing else. I've said that about noise, but never image quality. The OP asked about image quality, which I furthered narrowed as image detail. The first thing I said was that image detail is affected by lens aberrations. You can't take an image without a lens (unless you're into pinhole cameras.)

Graystar was also wrong earlier:

"Comparision of RAW images show that the G10, having nearly 50% more
pixel than the LX3, has pretty much the same noise performance as the
LX3. Obviously there's a problem with DxO's analysis."

On DxO's analysis...

Where do you see a reference to DxO Labs comparisons of the G10 and LX3 sensors in my statement? I was talking about MY OWN comparison of the G10 and LX3 RAW images, which I've written about in other posts. The analysis referred to in the last line is DxO Labs comparison of the 350D to the 1DsMIII, which was the topic of the entire post. Why would I rely on one DxO analysis to support my point that another DxO analysis isn't accurate?

You're taking my words out of context left and right!

Unlike others here, I don't care about the math. I only care about what I see in the final print. If two cameras produce identical prints with the same apparent detail levels and noise levels, then it doesn't matter what the graphs say...in my eyes the performance of the two cameras are equal.

Because I only care about the final print, any terms I throw out such as “superior detail” and “image detail” are all necessarily subjective. Here...I've provided my own definitions for my own terms...

“Superior detail” - When one print of a scene clearly exhibits greater image detail than another.

“image detail” - The recognizable detail in an image. Depending on the scene, a print with greater resolution may exhibit superior image detail, despite having more noise, than a lower resolution image with less noise. This occurs when signal levels are high enough (bright images) and contrast is high enough such that detail is still recognizable through the noise.

But image detail alone doesn't make a quality image. That's why I separate image detail from image quality. Image quality is a combination of high detail, low noise, and other factors. When comparing a G10 ISO 80 image to a 50D ISO 100 image, not only is the 50D's image detail much better than the G10, but the image quality is far better as well.

John Carson wrote:

Your interpretation of that passage makes no sense. What would be the
point of the new sensor design if it offers no advantage over the
old?

It does offer an advantage. As far as I know, the older sensor doesn't offer these enhanced low-resolution modes. So the new sensor offers two new features that can be applied when it is advantageous to do so.

The proper interpretation of the passage is that it performs as
well in favourable lighting conditions and better in more challenging
conditions.

It's hard to say it performs better when these features never existed in previous generations. What are you comparing these features to? Right now you can only make comparisons against images reduced in resolution during post-processing.

But really...if you want a 6MP camera you don't have to wait...just buy a Nikon D40 now.

Pixels...Quality.....
There is one more ingredient ie Print Size.

Print size must be taken into account. When you run out of pixels to get a required sized, quality print, then more pixels can be, and are a good thing.
jules

Diopter wrote:

One day I penned a piece about that. It is here:
http://www.dpreview.com/articles/compactcamerahighiso/

Funny, the article says Simon Joinson.

Anyway, it has some issues. The biggest one is that the logic used to say that since a tiny 10MP sensor is much smaller than a large 10MP sensor, that this necessitates smaller pixels for the smaller sensor, and therefore that is the reason for more noise. That is not true at all. A tiny 0.4MP P&S sensor with 10 micron pixels would have far more image noise than a FF sensor with 8 or even 1 micron pixels.