I have never discussed downsizing the image as method of noise reduction. What I say is that shot noise
mainly depends on sensor size and not pixel size, when the image is printed. I also assume that we are interested in printing large.
The original posting on this thread refers to an experiment I made: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/60-what-about-36mp-dslrs
The main advantages with smaller pixels I see is mainly:
- Aliasing effects are reduced
- OLP filtering is effectively reduced
- The image will need less sharpening and also respond better to sharpening, at least at low ISO
But I'm pretty sure that OLP-filtering is still needed. Aliasing would only be possible if the lens transfers significant MTF at pixel pitch and I'm surprised how well lenses hold up to modern sensors. On the NEX-7 "BJanes" measured 109 lp/mm on the DPreview comparison image. I have seen some artifacts and also color Moiré on my Sony Alpha 55 SLT. Sometimes it shows up as detail that is probably unnatural.
There has been some discussion on negative side effects of smaller pixels, with possible crosstalk between cells.
This article may be of interest regarding optimal pixel size: http://isl.stanford.edu/~abbas/group/papers_and_pub/pixelsize.pdf
Anyway the issue is complex. One aim is the best possible reproduction of detail, but that also needs careful work. Camera on tripod, MLU, exact focusing and using moderate apertures to avoid diffraction. The other extreme is low light shooting under free hand condition. Sensor FWC will be underutilized, and read noise much more visible.
It seems that the cameras of today having best high ISO performance have relatively large pixels, but don't have very good DR at low ISO, because of high amount of readout noise. It is also my understanding that high ISO cameras normally use pre amplifiers to achieve medium ISO values, but that means that the signal coming out of the sensor must be cleaner than the unamplified signal going into the ADC, otherwise amplification would not help. Why then is no preamplification used at base ISO?
Fair enough, but I have difficulty understanding why one gets involved with the drawbacks of smaller sensels (increased per-pixel noise, and larger file size) unless physically larger output with higher resolution is the (optional) goal, or as a means to reduce aliasing.
There are better methods for potentially reducing noise than downsampling.