Pixel Density vs. Noise vs. Resolving Power .... Does it translate?

It's been obvious that we cannot stop train of advancing pixel density to make

higher mega pixel cameras. This drives the marketing and eventually drives the

sales so we know the pixel count will go up.

 

For SLR's this was a good thing until they started hitting a noise limit where

increasing the pixel density has also increased the per pixel noise at high

ISO's. This limit seems to have been reached when going from the XT(350D,8MP)

to XTi(400D,10MP). Now with the announcement of XSi(450D,12MP), it seems that

the per pixel noise will inevitably increase again. This is the same situation

with 30D to 40D and for point and shoots(P&S) but for P&S's this limit was

reached a long time ago.

 

Many photographers would rather see better high ISO performance and no increase

in pixel count. Many would often consider this increasing per pixel noise as

backwards steps, which they are not wrong to do so.

 

What I suggest is a shift in the focus and use a metric other than per pixel

noise to judge if the sensor got better or not. What is important shouldn't be

the per pixel noise but the resolving power of the sensor. We gained per pixel

noise by moving up to higher MP sensors but we've also gained additional pixels

to give us more information. The higher pixel count may compensate for the

increased per pixel noise. We should take those into account to see if the

sensor truly took a step backwards.

 

In the end, what we should be looking for is if the sensor (at high or lower

ISO's) can resolve more detail than the previous generation of the sensor.

Maybe per pixel noise will go up, but in the end, if you end up with better

resolving power and able to see finer details at high ISO's then it is still a

step in the right direction.

 

Of course, the higher MP count also gives other trade offs like needing longer

processing times and larger storage space, but I don't see this as much of a

problem.

 

I give credit and thanks to those review sites that takes account of the

increased resolving power from higher MP count when they do their noise

analysis. I just hope that in the end, people are making their purchasing

decisions based on high ISO detail resolving power of the sensor and not from

just high ISO per pixel noise. "The answer is in the details."

 

Sincerely,

Weiyang Liu

"The higher pixel count may compensate for the increased per pixel noise. We should take those into account to see if the sensor truly took a step backwards." Another way of looking at it is to say that comparing 100% crops from sensors with the same size and different pixel counts does not answer the question we ought to be asking.

I'm not sure that the following actually plays out in tests of the sensors:

 

"This limit seems to have been reached when going from the XT(350D,8MP) to

XTi(400D,10MP). Now with the announcement of XSi(450D,12MP), it seems that the per

pixel noise will inevitably increase again"

 

There is not really a "limit" on noise levels per se. At one time it was, no doubt, difficult to

build a 6MP 1.6x crop sensor with sufficiently low noise. There has been progress - some

in the design of the sensor and some, no doubt, in the way that camera processes the

captured images.

 

The reports I read by and large stated that the 10MP sensors did not bring any detectible

increase in image noise compared to the 8MP sensors. While there have undoubtedly been

challenges in making it happen, it is not impossible that the 12MP crop sensors will

accomplish the same thing. Let's wait and see.

I think that Mr. Clark has made a reasonable attempt to quantify various aspects of sensor performance. See his results here:

 

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/index.html

 

However, I don't think it is sensible to swap one single absolute measure with another. The needs of different photographers for resolution, dynamic range of capture and low light performance vary - as was always clear when using film, where different emulsions and chemical processes cater to the different needs. This has become somewhat obscured in the digital age. Matters are further obscured by the impact of sensor optics (AA filters, microlenses and IR/UV cut filters) and image processing, both in and out of the camera and in the drivers for your display device or printer, as well as the technology of the final display. In the end it is the quality of the displayed image that counts, although the recorded image can be re-interpreted with better image processing and output in the future, placing a significant value on the quality of the original capture.

After looking at the data sheets for the latest Dalsa and Kodak sensors, the first thing that strikes me is the relative fill-factors of their sensors, and the photon efficiency. Kodak's sensors appear to be behind in both areas, so "pixel" density and size is no real guide to the actual performance of a sensor.

 

I'm also surprised to see that both companies are still using the illogical Bayer pattern for colour separation. Since the Bayer pattern uses twice as many green sensors as red and blue, it follows that the efficieny is reduced by almost 50% straight away, since the gain of the red and blue channels must be boosted twice as much as the green channel to acheive a roughly neutral colour balance.

 

Employing a "triad" sensor arrangement (viz the arrangement of phosphor dots on a CRT screen) makes far more sense; where the RGB pixel geometry is arranged and processed in groups of three, rather than four. This would enable an immediate increase in both true pixel density and light efficiency, thereby also reducing noise.

 

The efficiency of current RGB filter pigments is also quite low. The emphasis seems to be on making filters as narrow band as possible, with minimal spectral overlap. This gives us dazzling colour saturation, but does absolutely nothing for colour accuracy, since true monochromatic spectral colours can't be accurately represented.

 

If we are to stick to the stupid Bayer arrangement, it can be easily modified to give quads of Red, Yellow, Cyan and Blue. This would increase both colour accuracy and efficiency, while current pixel density and size would remain the same.

 

By the way Dalsa et al. This document constitutes verifiably dated prior publication. So don't try to patent these ideas eh?

At least from my limited experience I think Weiyang is right about the limits already being reached with P&Ss. My 7 megapixel and my wife's 8 megapixel P&S cameras have noticeably worse IQ than the 5 and 4 megapixel P&S cameras of 3 years ago.

 

I shudder to thing how bad the latest 10 and 12 megapixel P&Ss are.

 

We may not have yet reached the limit with APS-C sensors, but we may not be too far off.

I think there are still some tricks the manufacturers have up their sleeves in the race to higher megapixels while keeping overall image quality constant if not better.

 

http://www.dcviews.com/press/Sony-CMOS-back.htm

In addition to the issue of noise associated with increased pixel (photosite) density, thanks to the effects of diffraction, there is a correlation between pixel density and the number of f/stops you can use, from among those available on a given lens, without inhibiting a desired print resolution at an anticipated enlargement factor.

 

The f-Number at which diffraction will begin to inhibit a desired print resolution for an anticipated enlargement factor can be calculated as follows:

 

N = 1 / desired print resolution in lp/mm / enlargement factor / 0.00135383

 

An increase in the desired print resolution requires a smaller f-Number.

 

An increase in enlargement factor requires a smaller f-Number.

 

Conversely...

 

 

 

Use of a larger f-Number (smaller aperture) requires either a decrease in your desired print resolution and/or a decrease in enlargement factor. If you insist on using f-Numbers that are greater than the value calculated with this formula, you'll have to make due with a lower print resolution and/or a smaller print.) Such compromises are frequently necessary with high-density sensors (>400 pixels/mm), but seldom necessary with low-density sensors (<200 pixels/mm). Unfortunately, the vast majority of people working with high-density sensors are completely unaware of the fact that they are suffering a compromise in print resolution when they choose to make prints as large as the pixel count encourages and fail to select the widest aperture (smallest f-Number) provided by the manufacturer.

 

As pixel density increases, the number of f/stops that will support a desired print resolution at an anticipated enlargement factor will decrease, as will the number of corresponding shutter speeds available for your use. For example, if diffraction forces you to shoot wide open to support your desired print resolution at an anticipated enlargement factor, you’ll also have a choice of only one shutter speed at any given ISO setting.

 

- Pixel densities at or below 110 pixels/mm, will permit you to use nearly every f-Number available on any lens that's purpose-built for a given format (this excludes using large format lenses on a 35mm-sized sensor, for example), without inducing diffraction that would inhibit a desired print resolution of 5 lp/mm in a print that's scaled to the dimensions had at an unresampled image resolution of 360 dpi (necessary to support 5 lp/mm in a CMOS sensor that suffers losses due to the Bayer algorithm and AA filter).

 

This is true, whether we’re talking about a 5 MP sensor or a 50 MP sensor, a 7mm diagonal sensor or a 70mm diagonal sensor. As the pixel density increases, the number of stops that will support a desired print resolution of 5 lp/mm in a print that's scaled to the dimensions had at an unresampled image resolution of 360 dpi will decrease.

 

- At about 200 pixels/mm, roughly half of all the f-Numbers offered on a lens (and half of all the shutter speeds that would otherwise be available at a given ISO setting) will be incapable of supporting a desired print resolution of 5 lp/mm at the enlargement factor had when the print is scaled to a corresponding image resolution of 360 dpi (required to compensate Bayer and AA losses).

 

- At about 400 pixels/mm, whether you are using a 5 MP sensor or a 50 MP sensor, a 7mm diagonal sensor or a 70mm diagonal sensor, if the combination of pixel count and sensor dimensions yields a pixel density that high (>400), you'll find yourself limited to shooting wide open or nearly so, with only one or two corresponding shutter speeds from which to choose, if you are to prevent diffraction from inhibiting a desired print resolution of 5 lp/mm in a print that's scaled to the dimensions had at an unresampled image resolution of 360 dpi (necessary to support 5 lp/mm in a CMOS sensor that suffers losses due to the Bayer algorithm and AA filter).

 

- At 600 pixels/mm and greater, (as with the extremely dense sensor found in the Pentax Optio A20 and A30, for example, which has a density of 636 pixels/mm), you will find yourself with absolutely NO f-Numbers capable of supporting a desired print resolution of 5 lp/mm at the enlargement factor had when the print is scaled to a corresponding image resolution of 360 dpi (required to compensate Bayer and AA losses). The lens simply won't be fast enough to prevent diffraction from inhibiting that resolution at that enlargement factor.

 

Even if your personal choice of a desired print resolution is only 1 lp/mm (equivalent to an image resolution of only 72 dpi, taking Bayer and AA losses into account for CMOS sensors) the fact remains that any increase in pixel density (whether we've reduced sensor dimensions or have increased pixel count) will have a corresponding decrease in the number of stops that will support your desired print resolution of 1 lp/mm. With high-density sensors, you are welcome to use all the stops made available on the lens, but you'll have to reduce either your desired print resolution or the enlargement factor.

 

Mike Davis

http://www.AccessZ.com