MP/ISO

[recon photography]

many people are saying they want lower mega pixels to combat noise performance, my understanding is that the more mega pixels on a sensor the more noise there is, per pixel and so there will be no positive effect on noise performance by reducing mega pixels on the overall image as a whole, is this correct?

i have in fact read that the more megapixels the finer the grain is, which would create a smoother nicer effect on the whole image

is this information incorrect? any1 have any good websites explaining this?

thanks

 

[recon photography]

why doesn't this show up on forum pg :O?

 

[Loswr]

why doesn't this show up on forum pg :O?

The list of recent posts on the CR home page updates every hour or so, and with the frequent posts to the Contest pages, many non-Contest posts just don't show up sometimes. Your post does show up in the forum pages - that's how I found it...

Ps. it's there, now.

my understanding is that the more mega pixels on a sensor the more noise there is

The idea that smaller pixels mean more noise is commonly held, but in fact, it's wrong. If you take a single pixel in isolation, it's certainly true that a smaller pixel has lower full-well capacity, less sensitivity, and more noise. But people then assume that means that a whole sensor composed of smaller pixels will have less sensitivity and more noise than a sensor with larger pixels, and that is the false assumption. Noise scales directly with the spatial frequency (i.e. the level of detail in an image), and inversely with the total amount of light falling on the sensor. There are a myriad of reasons that people think smaller pixels equate to more noise, all based on incorrect assumptions (e.g. that you can compare 100% crops from sensors with different pixel densities to judge noise, which is like comparing film grain with a 2x loupe vs. a 5x loupe), or cases where correlation does not mean causation (e.g. P&S cameras have more noise, they have smaller pixels, therefore smaller pixels mean more noise - in that case, it's the smaller sensor, not the smaller pixels).

Processing algorithms matter, too. For example with long exposures Nikon automatically applies a median blur filter to reduce noise in the RAW image, even with long exposure NR turned off, i.e. partial processing of the RAW image in-camera, in a way that cannot be turned off by the user (the workaround for astrophotographers shooting Nikon is 'mode 3' - set long exposure NR to on, but power off the camera during the dark frame acquisition).

Bottom line - sensor size has the largest impact on image noise. Reducing the MP count of the sensor will not substantially reduce the noise.

Before someone launches into a comparison of ISO noise on the FF Canon 5DII vs. the FF Canon D700, claiming that the lower MP count of the D700 results in its lower noise, remember that spatial frequency matters. Yes, the D700 has less noise, but that's not because of the larger pixels per se, it's because the 5DII has a higher spatial frequency (more detail). If you downrez a 5DII image to 12 MP, it will appear to be much less noisy (it won't really be less noisy, but it will be perceived that way) - so, you'll get something that looks a lot like a D700 image, in terms of noise and detail. That doesn't make the 5DII or the D700 'better' - they are simply different, reflecting different emphases by the two manufacturers.

Under most circumstances, I prefer Canon's approach - starting with a higher resolution and noisier mage, I can trade a loss in detail for more noise reduction; starting with a lower resolution but cleaner image, I can't trade an increase in noise to add detail that's not present in the original image.

 

[Rocky]

Before someone launches into a comparison of ISO noise on the FF Canon 5DII vs. the FF Canon D700, claiming that the lower MP count of the D700 results in its lower noise, remember that spatial frequency matters. Yes, the D700 has less noise, but that's not because of the larger pixels per se, it's because the 5DII has a higher spatial frequency (more detail). If you downrez a 5DII image to 12 MP, it will appear to be much less noisy (it won't really be less noisy, but it will be perceived that way) - so, you'll get something that looks a lot like a D700 image, in terms of noise and detail. That doesn't make the 5DII or the D700 'better' - they are simply different, reflecting different emphases by the two manufacturers.

Under most circumstances, I prefer Canon's approach - starting with a higher resolution and noisier mage, I can trade a loss in detail for more noise reduction; starting with a lower resolution but cleaner image, I can't trade an increase in noise to add detail that's not present in the original image.

That is the best explaination of MP vs Noise explanation that I have seen.

 

[Leopard Lupus]

Neuro takes the cake with this explanation. Well done, sir. 1+

 

[Meh]

Neuro, are you sure it's rigorously correct to say "smaller sensors have more noise regardless of pixel size" (I'm paraphrasing your statements somewhat), "smaller pixels have less sensitivity" (define 'sensitivity'), and that "reducing MP will not substantially reduce noise"?

Take a case of the same camera, same lens but test two sensors of different sizes but same design, same spatial resolution (pixel pitch/size). The noise is the same, to say otherwise is suggesting that the noise in the centre of any given sensor is higher than that around the edges. The same image (i.e. same composition) taken by a smaller sensor has to be enlarged more to get any given viewing size and the image will not look as good because there would be less resolution (assuming viewing size is large enough to see the difference) but that is not due to noise in the image... it is due to less resolution which is not the same thing.

Now consider two sensors, one larger than the other, with the same number of pixels. Camera and lens are still the same. The smaller sensor has more noise, but not because the "sensor is smaller" per the above but rather because the photosites are smaller and each collects less photons therefore the SNR is higher. So again, compose each image the same, the frame from the smaller sensor is enlarged more to get the same viewing size but now each image has the same number of pixels. The image from the smaller sensor has more noise because there is more variation (i.e. noise) from pixel to pixel.

So what about noise-reduction algorithms. They work better the more information (resolution) there is to work with but megapixels are megapixels. It doesn't matter whether the data (say 18MP) comes from a FF, APS-C, or smaller sensor. The algorithms work on "data" and it's irrelevant whether that data came from a larger or smaller sensor. Again, I'm taking the case here the same scene is imaged on the sensor regardless of size. The NR algorithms work better but the image starts with more noise... the final image from the sensor with smaller photosites is still noisier and in addition has the NR softening so overall the image quality is worse.

In the case of P&S there is in fact another difference to consider. The lens is not the same. It is smaller, has a shorter focal length, smaller aperture (actual size), and usually lower quality than DSLR lenses which in most cases is a significant issue. For any given f-ratio the aperture is in fact much smaller and that causes greater diffraction AND the higher spatial resolution of the sensor will pick that up to a greater degree. This is offset somewhat because the image size at the focal plane is smaller but not entirely. However, this is not "noise" in the context of what we're talking about and results in a general softening of the image rather than visible variation.

A good distinction to make here for folks who might not realise, is that the noise we see in any average to well-lit scene is primarily due to photon noise not due to the electronic noise (read noise) that most people think about when talking about noise. Photon noise actually increases with increased brightness but only increases as the square root of the number of photons absorbed by the sensor so the SNR of brighter scenes is higher than darker scenes. Read noise is relatively stable and does not increase with brightness (an exception is blooming but that is well controlled in modern sensors) therefore the SNR due to the read noise is really insignificant for reasonably bright scenes. The fact that SNR (from photon and read noise) improves with the number of photons absorbed is a valid reason for "expose to the right" although I've read another reason that I don't believe is correct.

If any of the above isn't correct, what am I missing?

 

[Orangutan]

I'd like to offer a hypothesis that might explain both sides of this disagreement: when it comes to tiny electronics, smaller structures are harder and more expensive to make than larger structures. I'm willing to accept what Neuro (and others previously) have explained regarding overall sensor size. Nevertheless, I find it difficult to believe that, at the same price point, a small-photosite sensor can be made as "clean" as a large-photosite sensor. In short, I think it's a manufacturing issue, not a theoretical issue. I'd believe that you could make the two equal if you accept a higher per-unit cost on the small-photosite sensor. It would be nice to hear from EE's who've worked on related tech.

I'd love to have a high-MP, low-noise sensor, but I doubt I could afford it. For now, 18MP is enough for me, but I still want lower noise.

The idea that smaller pixels mean more noise is commonly held, but in fact, it's wrong. If you take a single pixel in isolation, it's certainly true that a smaller pixel has lower full-well capacity, less sensitivity, and more noise. But people then assume that means that a whole sensor composed of smaller pixels will have less sensitivity and more noise than a sensor with larger pixels, and that is the false assumption.

 

[archangelrichard]

There is another issue you are ignoring - the more MP in the same area the more important the proximity of the signals interfering with each other becomes.

you have to go to lower power (and subsequently less sensitivity) to control this which just isn't practical; this is why a larger (FullFrame) sensor is less noisy at higher ISO's (higher amplification) - there is more distance between the lines so they offer less interference to one another

This is basic physics

 

[caruser]

The idea that smaller pixels mean more noise is commonly held, but in fact, it's wrong. If you take a single pixel in isolation, it's certainly true that a smaller pixel has lower full-well capacity, less sensitivity, and more noise.

I'm with you on the noise, but at some point the smaller per-pixel full-well capacity must start to dominate by reducing the possible dynamic range per-pixel, which in this case translates into reduced dynamic range for the whole image (like "of what use is a 14bit A/D if you can only hold 400 photons per pixel")?

 

[NotABunny]

You'll see a lot of people who come up with all sorts of theoretical arguments that smaller pixels equals noisier images. The trouble is that when physical proof is asked, they either only show photos taken in different technical conditions (the absolute amount of light clearly alters the noise, just experiment a bit), or don't know that in order to understand if the sensor size or the pixel size actually affect the noise, both the sensor size and resolution must be normalized to the same unit area (like square millimeter) and only then you can look at the (entire) photos (which are then viewed normalized at display size).

It appears that sensor manufactures have, for a given used technology, a certain maximum resolution up to which they can go without increasing the noise levels per image.

Would it be possible for them to use the same (current) technology but with a decreased resolution in order to output images with lower noise levels? Unlikely. The fact is that they have so far either preserved or lowered the noise levels per image while increasing resolution.

They may be able to use a different technology with a smaller resolution in order to output images with lower noise levels, but the fact that they don't do it shows (to me) that such technology is simply too expensive for mass products.

Also take into consideration that Canon has a sensor that's over 20 * 20 cm. This should prove to you that decreasing the resolution doesn't have a visible effect on the noise levels per image (or, if it has any, it's marginal). With the technology they have, the only way to visibly lower the noise is to increase the sensor + lens size (since that physically captures more light for the same exposure).

There are however tricks that may be used (for mass products) to lower the noise levels, like multiple readings of the electrical charge, either at the same ISO or at different ISOs in order to average the readings.

Given various rumors, I hope Canon will bring something like this with their next camera.

If the next Canon pro-camera will have a resolution lower than expected, something like 16...20 MP rather than 30...40 MP, it will not be because larger pixels equal lower noise levels per image, but because they use tricks like the multiple ISO readings (in which case the pixel size may start to have a visible effect on the per image noise levels).

Some people might say that such a low resolution might be due to the video optimizations, but I think that this resolution is still too high to be dismissed this easily. Maybe if the camera were to have 8...10 MP, I would agree (since this would mean a 4K sensor read without line skipping).

 

[Bob Howland]

Bottom line - sensor size has the largest impact on image noise. Reducing the MP count of the sensor will not substantially reduce the noise.

However, at the highest ISO's (i.e., amplifier gains), there are so few photons per pixel that the baseline noise in the sensor, amplifier change and A-to-D converter becomes more visible. That is when reducing the MP count, i.e., using larger pixels, for a given size sensor becomes rational.

Edit: Should be "amplifier chain", not "amplifier change"

 

[Bob Howland]

Also take into consideration that Canon has a sensor that's over 20 * 20 cm. This should prove to you that decreasing the resolution doesn't have a visible effect on the noise levels per image (or, if it has any, it's marginal). With the technology they have, the only way to visibly lower the noise is to increase the sensor + lens size (since that physically captures more light for the same exposure).

And Canon reportedly has a 40MP FF sensor that they decided not to use because of poor noise characteristics at high ISO. Increasing sensor size may be good, but what happens when the realities of the marketplace impose a maximum size of 24mm x 36mm?

 

[NormanBates]

I have absolutely no hope of leading this neverending debate to an end (becase it is, you know, neverending), but I actually ran some empirical tests a year ago:
http://www.similaar.com/foto/mpix/mpix.html

based on those results, I'm on the "more megapixels means more noise per pixel, but not necessarily more noise per image" camp

also, dxomark tests noise levels "per image", not "per pixel" and arrives to similar conclusions: per-image snr on 7D (18mpix APS-C, released nov-2009) and D300s (12mpix APS-C, released nov 2009) is about the same, with the 7D actually having an advantage at high ISO values:
http://www.dxomark.com/index.php/Cameras/Compare-Camera-Sensors/Compare-cameras-side-by-side/%28appareil1%29/619|0/%28brand%29/Canon/%28appareil2%29/614|0/%28brand2%29/Nikon

 

[Loswr]

Great points, everyone!

Take a case of the same camera, same lens but test two sensors of different sizes but same design, same spatial resolution (pixel pitch/size). The noise is the same, to say otherwise is suggesting that the noise in the centre of any given sensor is higher than that around the edges. ...

Now consider two sensors, one larger than the other, with the same number of pixels. Camera and lens are still the same. The smaller sensor has more noise, but not because the "sensor is smaller" per the above but rather because the photosites are smaller and each collects less photons therefore the SNR is higher. ...

So what about noise-reduction algorithms. They work better the more information (resolution) there is to work with but megapixels are megapixels. It doesn't matter whether the data (say 18MP) comes from a FF, APS-C, or smaller sensor. The algorithms work on "data" and it's irrelevant whether that data came from a larger or smaller sensor. Again, I'm taking the case here the same scene is imaged on the sensor regardless of size. The NR algorithms work better but the image starts with more noise... the final image from the sensor with smaller photosites is still noisier and in addition has the NR softening so overall the image quality is worse.

The missing factor is one discussed above, as highlighted in bold by NotABunny. What we generally care about is noise in the image, not noise from each pixel. To the point in your above examples, the missing factor is that with a smaller sensor, you have less total light hitting the sensor. The problem, in part, is that most ways of measuring sensor performance do so at the pixel level - that's how sensor performance is assessed in the QC setting, for example. So, sensitivity is measured in photoelectrons/lux second/pixel. Read noise is measured in RMS electrons/pixel or ADU/pixel. Dynamic range is measured in stops/pixel or dB/pixel, etc. All of those measurements ignore the total image, and the spatial fraquency of the sensor - and that's what determines image noise (which is what we really care about).

In terms of noise reduction algorithms, that's an interesting point. As far as I know, the in-camera NR processing is done only at the pixel level, i.e. each photosite is treated independently and processed as such. When NR is done in post-processing, empirically, some software does a better job than others, e.g. DxO, NoiseNinja and Topaz Denoise all do a better job at reducing noise while maintaining sharpness than Canon's DPP. I wonder if that's because like the in-camera NR, DPP is doing NR at the pixel level (albeit with more powerful algorightms), while the superior NR programs are using some sort of nearest neighbor analysis to reduce noise of individual pixels based on signal in surrounding pixels.

A good distinction to make here for folks who might not realise, is that the noise we see in any average to well-lit scene is primarily due to photon noise not due to the electronic noise (read noise) that most people think about when talking about noise. Photon noise actually increases with increased brightness but only increases as the square root of the number of photons absorbed by the sensor so the SNR of brighter scenes is higher than darker scenes.

This makes sense, as well.

It also raises another issue - how the ISO noise tests are commonly done. If you're shooting a test scene/chart and successively increasing the ISO, you have to decrease something else in parallel to keep the exposure the same. You can vary the aperture, but then you're not shooting the same scene because the DoF is changing (and OOF areas subjectively appear to have more noise). So, if you keep the aperture the same so you're taking the same picture, then you're decreasing shutter speed to compensate for the increased ISO - and that minimizes the contribution of shot noise. I don't really think tests done like that are fair, since they don't represent real-world use (i.e. it's not common to use ISO 3200 and 1/4000 s shutter speed - if you have that much light, you shoot at a much lower ISO, right?).

Like NormanBates, I've done some empirical testing, in this case during the course of comparing the 7D and 5DII. When I used a constant f/8 and varied shutter speed in tandem with ISO, from 1/30 s at ISO 100 to 1/8000 s at ISO 25600, the noise certainly increased with increasing ISO, but honestly it didn't look as bad as I thought. But as I stated, would anyone intentionally shoot a real-world shot at ISO 25600 and 1/8000 s? When I took a different approach, keeping aperture and shutter speed constant and knocking down the illumination a stop at a time with ND filters (less light, just like when you'd use high ISO in the real world), the ISO noise got worse much faster. To me, the noise in the ISO 25600 1/8000 s shot looks fairly similar to that in an ISO 6400 1/60 s shot (with 6-stops of ND to simulate dim lighting).

based on those results, I'm on the "more megapixels means more noise per pixel, but not necessarily more noise per image" camp

So am I, although I'd state it as "smaller pixels means more noise per pixel, but not necessarily more noise per image."

 

[akiskev]

As far as I know, the in-camera NR processing is done only at the pixel level, i.e. each photosite is treated independently and processed as such.

The simplest nonlinear noise reduction algorithm replaces the central coordinate of each embedding vector by the local average of this coordinate.
In other words there is no such thing as independent treatment of each pixel in noise reduction. It doesn't make sense. A comparison is always needed.

 

[epsiloneri]

Reducing the MP count of the sensor will not substantially reduce the noise.

This is correct to first order, and is usually the case for well-exposed images. For under-exposed parts of an image where read-out noise becomes important, fewer pixels show less noise (because read-out noise is typically similar per pixel independent of size).

However, at the highest ISO's (i.e., amplifier gains), there are so few photons per pixel that the baseline noise in the sensor, amplifier change and A-to-D converter becomes more visible. That is when reducing the MP count, i.e., using larger pixels, for a given size sensor becomes rational.

Normally it's the read-out noise that matters, discretisation noise per pixel is offset by the fact that you add more smaller pixels together for the same solid angle as a bigger one.

I'm with you on the noise, but at some point the smaller per-pixel full-well capacity must start to dominate by reducing the possible dynamic range per-pixel, which in this case translates into reduced dynamic range for the whole image (like "of what use is a 14bit A/D if you can only hold 400 photons per pixel")?

You would think this is the case, but the full-well capacity usually scales with area, so the dynamic range per solid angle does not depend on pixel size. In other words, smaller pixels have smaller wells, but since you add up more of them, this (almost) exactly adds up to the well of the larger pixel with the same total area.

 

[caruser]

I'm with you on the noise, but at some point the smaller per-pixel full-well capacity must start to dominate by reducing the possible dynamic range per-pixel, which in this case translates into reduced dynamic range for the whole image (like "of what use is a 14bit A/D if you can only hold 400 photons per pixel")?

You would think this is the case, but the full-well capacity usually scales with area, so the dynamic range per solid angle does not depend on pixel size. In other words, smaller pixels have smaller wells, but since you add up more of them, this (almost) exactly adds up to the well of the larger pixel with the same total area.

Sorry, I'm lost, what's this "solid angle"?

Anyhow, what you are saying is that the larger pixel's larger well doesn't help because if the well is Z times larger it also collects Z times the light and therefore saturates equally fast?

Sounds so plausible that I wonder why I didn't think of it myself ;-)

 

[Loswr]

Anyhow, what you are saying is that the larger pixel's larger well doesn't help because if the well is Z times larger it also collects Z times the light and therefore saturates equally fast?

Correct, for the most part. You take 4 smaller pixels that fit into the area of one larger pixel, and while the larger pixel has a deeper well, it's getting more light, and if you sum the wells of the smaller pixels, they're equivalent to the single larger pixel, and together they get the same amount of light.

 

[caruser]

Anyhow, what you are saying is that the larger pixel's larger well doesn't help because if the well is Z times larger it also collects Z times the light and therefore saturates equally fast?

Correct, for the most part. You take 4 smaller pixels that fit into the area of one larger pixel, and while the larger pixel has a deeper well, it's getting more light, and if you sum the wells of the smaller pixels, they're equivalent to the single larger pixel, and together they get the same amount of light.

Thanks, and what about the other issue, is that a red herring too? If say a small pixel can capture only 500 photons you'll only ever get 9 bits out of each one, but if you have one quad-size pixel you'll get up to 2000 photons or 11 bits? Don't the smaller pixels limit dynamic range here? Or is the gain with the larger pixel non-existent because you only gain two more noisy lower bits, rather than two higher bits beyond the 9 bits?

 

[epsiloneri]

If say a small pixel can capture only 500 photons you'll only ever get 9 bits out of each one, but if you have one quad-size pixel you'll get up to 2000 photons or 11 bits? Don't the smaller pixels limit dynamic range here? Or is the gain with the larger pixel non-existent because you only gain two more noisy lower bits, rather than two higher bits beyond the 9 bits?

If you combine 4 pixels of 9 bits each the combined DR is precisely 11 bits, I think that answers your question

BTW, "solid angle" is just a 2D-angle (see the wikipedia explanation)