I always think about sensors pixels as small buckets catching light particles. After one measurement cycle, the number of particles in each bucket are counted and that's the raw pixel value. A bucket has a ceratin contents. If it's full and particles keep falling in, they will spill over and not be measured (these are highlights being blown out). Also, there's some ennoying drizzle of light particles that's continuously there (this is the noise). At low light level, only few particles fall in the bucket, and the drizzle particles will mask the actual light signal you're measuring.
Now how to reduce the impact of the drizzle particles on the final raw pixel values? There is only one way: increase the exposure. But then the highlights will be blown out! So now we have to rescue the highlights. I wonder if it is feasible to look into the buckets during exposure and if a bucket fills, very quickly empty it and keep track of this. Then after a complete exposure, just add 4096 (if it's a 12 bit sensor) to the raw pixel value of the emptied buckets and you've increased the sensor's dynamic range by one stop, while the noise level stays the same. You could even imagine repeating this procedure multiple times. Another approach may be to measure how quickly a particular bucket fills and extrapolate this measurement to get the number of particles that would have hit the bucket if it would have been large enough (and assuming that light intensity on each pixel stays the same during the exposure---may cause strange effects on moving subjects). But still, this most probably involves new sensor technology, and we're not looking for that, are we? Groeten, Vic >----- Oorspronkelijk bericht ----- >Van: Toralf Lund [mailto:[EMAIL PROTECTED] >Verzonden: woensdag, augustus 16, 2006 02:59 PM >Aan: 'Pentax-Discuss Mail List' >Onderwerp: Re: Holy Crap -- Pentax 10MP body > >Mark Roberts wrote: >> DagT wrote: >> >> >>> In my view the reduced noise is equivalent to increase dynamic range. >>> >> >> Not necessarily equivalent, but reduced noise certainly could be the >> *result* of increased dynamic range: Increased bit depth could >> conceivably facilitate the use of more advanced noise reduction >> algorithms that do more noise reduction with less damage to fine >> detail in the image. >> >No, I think he refers to an actual reduction of noise at the >signal-level, not a removal of "noise" from the digital data, and sees >this the other way around (more or less.) Less noise means that you can >actually use a larger subset of the range of charges in the sensor, so >as to speak. > >See >http://micro.magnet.fsu.edu/primer/digitalimaging/concepts/dynamicrange.html > >- Toralf > > >-- >PDML Pentax-Discuss Mail List >PDML@pdml.net >http://pdml.net/mailman/listinfo/pdml_pdml.net > > -- PDML Pentax-Discuss Mail List PDML@pdml.net http://pdml.net/mailman/listinfo/pdml_pdml.net
