Bits are not stops.

12-bits amounts to 2E12 gradations of R, G, or B. That's a lot, no?

Over a 6-stop range, the digital record has that many gradations possible.

Maybe someone else can explain it better.

I meant the LINEAR curve of the sensor.  Sorry.

I just read your "Expose to the right" with great interest.  I am thinking about upgrading to a DSLR with 12 bit A/D converter.  I have wanted one ever since a job of mine showed me what could be done with satellite images and and tonal curves.  

As an old computer scientist, I understand about the exponentially decreasing number of levels as you move to a lower level signal.

What hit me between the eyes was your statement that an SLR has 5 or 6 stops of range.  Whoa -- 12 bit A/D and 6 significant bits of data.

I can think of several possible explanations:
1) The article considers darker zones unusable because of significant "banding" from few intermediate levels.  Here a 12 bit converter should push that problem several zones further away. (In this case I should go ahead with my upgrade.)

2) We have a triumph of marketing over engineering. Perhaps, consumer sensors can't really produce a true 12 bit image.  You get a 12 bit value but the low order 6 are just noise.  (In this case, I should save my money for something else.)

3) Most likely.  The article meant something else by this comment and I just haven't figured it out yet.

Please help me understand.  In the right hand half of the historgram, how many significant bits should I be able to expect from a camera like a Nikon D70 or Digial Rebel?  

I really like to know before I buy a camera that might only disappoint me.

Thanks

Thanks for answering, it sounds like you are talking about the banding (option 1) because of inadequate gradations.  

As I understand it (see Noram Koren's site) the eye can see about 70 gradations per stop.  Thus, in 12 stops (12*70=840).  This is far less than the 4096 in 12 bits.  The rub is that the eye's curve and the exponential curve of the sensor don't line up.  This is the reason for exposing to the right--if we increase the exposure by a stop and then correct in PS, we get another stop of range before banding occurs.

I hope that is what the 6 stop limit was.  If it is the signal to noise threshold, then its an entirely different game.

Quote (Guest @ April 14 2004,16:34)

What hit me between the eyes was your statement that an SLR has 5 or 6 stops of range.  Whoa -- 12 bit A/D and 6 significant bits of data. I can think of several possible explanations: 1) The article considers darker zones unusable because of significant "banding" from few intermediate levels.  Here a 12 bit converter should push that problem several zones further away. (In this case I should go ahead with my upgrade.) 2) We have a triumph of marketing over engineering. Perhaps, consumer sensors can't really produce a true 12 bit image.  You get a 12 bit value but the low order 6 are just noise.  (In this case, I should save my money for something else.) 3) Most likely.  The article meant something else by this comment and I just haven't figured it out yet.

A DSLR has 12 bits of sensor data. Roughly 10 of those are usable. Since the sensor is linear in response, that means there is usable data over 10 stops of light.
This is 10 stops of range to some people.
To other people this is 5-6 stops.

I don't know for sure why this is so, but I did observe when testing my camera with pictures of a blank white wall and counted range as the difference between the brightest and darkest shots that showed ALL the texture of the wall, I got a figure around 4 stops lower than the number of usable bits. I attribute this to the wall actually having several stops of brightness difference even though the whole wall looked white to the eye.

What this means to me is a DSLR with 10 usable bits will capture all detail from zones 3 to 7, most detail in zones 2 and 8, and some detail in zones 1 and 9. You can call that a range of 5-6 stops, or 9-10 stops, however you want.