On Sun, Feb 24, 2008 at 11:05:04PM +0800, Sandy Harris wrote: > On Fri, Feb 1, 2008 at 2:51 PM, John Francis <[EMAIL PROTECTED]> wrote: > > > > So given, say, a K20D with its 14 Mp, can we somehow > > > combine sets of four dots to get a 3.5 Mp image with > > > better performance in "available darkness"? Or would > > > this also push noise up, perhaps to awfuI levels? > > > > > > How much would you gain? Four times the pixel area, > > > so in theory two stops, but would that happen in > > > practice? > > > > That's not how it works. When you add four pixels, each with > > a random noise component, you only improve the signal-to-noise > > ratio by a factor of two, not by a factor of four. So theory > > says the best you could hope for is one stop of improvement. > > That's counter-intuitive to me. You've got four times the signal > and four times the noise, so I cannot see why S/N ratio should > change at all. Can you explain the theory or point to a reference?
It's pretty simple. While the signal is correlated, so adding two pixels worth of signal gives you twice the value, the noise isn't. When you add the (random) noise from two sample sites the values may add, or they may cancel each other out. For more details try any basic statistics textbook. -- PDML Pentax-Discuss Mail List PDML@pdml.net http://pdml.net/mailman/listinfo/pdml_pdml.net to UNSUBSCRIBE from the PDML, please visit the link directly above and follow the directions.
