Monday, April 12, 2010

Sensor Size Versus Quantum Efficiency

FTF member GrahamD posted a nice table today, showing the sensor size relationship to stops of efficiency. All things being equal, one can estimate image quality (say, for noise characteristrics but that’s probably a little simplistic) in general terms using this table.

sensor_quantum_efficiency

The rather cool thing here is how close this comes to my empirical tests. Take the F80EXR at 12mp and the D700 at 12mp, all things being mostly equal (design of sensors is different, but both are more advanced than traditional CCD.)

The D700 is 35mm size, which is the baseline, hence 0 stops from baseline. The F80EXR is 1/2” sensor, which is 4.8 stops down in efficiency. Thus, we can expect approximately 5 stops difference in image quality. Yes, I’m being simplistic again, but bear with me.

If you look closely at the table, you will find that doubling of the sensor area will double the efficiency … which means increasing it by 1 stop. Makes sense, since you are gathering twice the light. And research by people at places like Clarkvision and DXO Labs has shown that pixel density is less of a factor than overall sensor size.

So the 5 stops should pretty much stand. And in my recent article showing the ISO ladder for the D700, that came really did come out about 5 stops ahead of the F80EXR.

But pixel density does come into play when camera software gets involved. For example, the higher pixel density of the F80 over the F70 has caused a severe drop in high ISO image quality as the light gets bad. A lot of chroma noise is pushed into the images and this is extremely difficult to remove.

You will end up shooting black and white in bad light with the F80, just as you must with the ZS3, because both emphasize the blur channel to an extreme, ruining the images.

But with the F70, the density must be below some threshold, because theit is higher edge detail at 1600 ISO and there is no chroma noise problem to speak of. Some subtle blotching does occur in bad light, but it is far easier to deal with than the disaster that is the F80 at 1600 ISO.

One further thing … the comparison between small sensor cameras can get rather vociferous at times, as can the comparison between dSLRs and small sensor cameras. We pretty much know that the small sensor cameras that are shipping today all fall into the range of 2/3” sensors (S100fs was the last of these, there are no more of these shipping today) and 1/2.5” sensors (typical long lens compacts and mini bridge cams) … with the 1/2.3” BiCMOS cams stuck near the bottom but claiming a massive increase in efficiency. That part is somewhat true, but shows up only at high ISO for some reason … at low ISO these cams seem to make quite the hash of details.

Anyway, back to the story. The point I am trying to make here is that the QE range from 2/3” to 1/2.5” is –3.9 through –5.1 … a 1.2 stop range between the best and the worst of small sensor cams. I.e. they are all pretty similar, centered at 4.5 stops down.

And they are all somewhere around 3.5 stops worse than the APS-C consumer dSLRs based on QE. So maybe it is time to put the silly arguments to rest. As pixel density rises, jpeg engines are forced to turn anything with low contrast to mush to control noise. And that shows very clearly in my D700 / F80 / F70 / ZS3 comparison.

No comments: