> Fra: Toralf Lund <[EMAIL PROTECTED]> > > > > Yes, you have to distribute your bits between the noise level and the > > saturation level. Some things, like an > >increase in the light sensitive area of each pixel or other tricks like > >reducing dark current, can suppress noise > >in one ends and other tricks like Fuji does with their sensors may give > >something in the other end.
> Wouldn't an increase in the light sensitive area essentially mean > increasing the native ISO? I mean, the absolute noise and even noise > relative to saturation level would be the same, but saturation would be > reached faster. I'm not sure how close those two follow each other, but you may be right. I think Fuji uses a smaller pixel to measure the highlights, which means that it does not reach saturation as fast as the larger pixels. Other things may be done, at least in theory, like increasing the drain from the overexposed pixel to avoid saturation, or making the drain nonlinear. > > In addition you can gain a little bit in how you distribute your bits, > > and maybe that is what they do in the 22bit > >conversion. > > > It's been mentioned before, but based on some info Rob Studdert dug up, > I'm lead to believe that there aren't any extra tricks involved in those > 22bits. It's probably all about a circuit designed to work with > different types of sensors, that *for internal use* converts the signal > from the sensor to a digital value with so many bits that you can be > pretty sure no information (not even the noise) is lost, no matter what > sensor you throw at it - and probably also so that calculations that > involve multiple steps won't loose accuracy between the steps (think of > what happens if you do a division followed by a multiplication on a > digital value.) It will do various types of gain/offset adjustments in > this domain rather than on the analogue signal so as to make sure no > *extra* noise is introduced (the original noise may still be amplified, > of course.) Somebody mentioned the use of curves to increase the nice parts of the signal and reduce the noisy part. That could be done with the 22bit conversion, but this is speculation. I often use curves that way in 16bit TIFF before I reduce it to 8bit jpeg. DagT -- PDML Pentax-Discuss Mail List [email protected] http://pdml.net/mailman/listinfo/pdml_pdml.net
