5D artifacts

[John_Sheehy]

[ QUOTE ]
I fail to be 'blown away' by DMR images - good though they are and I'm still not convinced 16bit is the magic ingredient. I regularly use a 22mp 16bit MF back (Sinar) - yes it has more resolution and more dynamic range ( nothing to do with 16 bit)

[/ QUOTE ]

There is a limit to how much DR you can have with 12 bits.

I will never understand why incorrect anecdotes propagate like wildfire, while true ones stagnate and die. The common notion that "Dynamic range has nothing to do with bit depth" is pure nonsense. Bit depth is one of the factors that defines dynamic range. Bit depth puts a limit on the dynamic range of each pixel. You can't have a dynamic range greater than the bit depth, at the pixel level, no matter how much or how little noise you have. If there was no noise at all, and you have a pure, idealistic analog capture, the dynamic range would be *COMPLETELY* determined by the bit depth, and nothing else. In the real world, there is noise, and noise reduces the DR of the individual pixel, but for a fixed level of digitization, a little noise *increases* the dynamic range of the image as a whole, at a lower spatial resolution, and this is optimum when the range of noise is one digitized level, and is evenly distributed (sensor noise is not).

The chance that the Leica back would be just as good with 12 bits is pretty slim. Most current large-sensor technology is limited by 12-bit digitization; not by noise. If this were not true, their high-ISO images would less detailed and more noisy than their low ISO shots under-exposed with the same shutter speed and aperture, and they are clearly not.

[John_Sheehy]

I wrote:

[ QUOTE ]
The chance that the Leica back would be just as good with 12 bits is pretty slim. Most current large-sensor technology is limited by 12-bit digitization; not by noise. If this were not true, their high-ISO images would less detailed and more noisy than their low ISO shots under-exposed with the same shutter speed and aperture, and they are clearly not.

[/ QUOTE ]

Sorry, I botched that last sentence up. I was going to write "their high-ISO images would be less detailed and more noisy than they are", and then I added the bit about comparing to under-exposed low-ISO shots, and grafted the two sentences together.

[John_Richard]

Quote
"There is a limit to how much DR you can have with 12 bits" - please explain exactly what this limit is then because I think it it is still way beyond what the actual chip can deliver in terms of it's S/N ratio and that is why the call for 16bit with current chips is a red herring.

Quote
"The chance that the Leica back would be just as good with 12 bits is pretty slim" - the qualities of the DMR are not just dynamic range - rendering of fine detail, tonal qualities and color are not directly a result of 4 extra bits. The AD convertor, the chip design (CCD vs CMOS), the lack of an AA filter and the superb leitz glass will have a far greater effect on the look of DMR images. If Canon or Nikon could produce much better images from their chips with 16bit then I think they would be doing it by now. As it is the dynamic range of the 5D, the 1Ds2 and even the 20D is pretty impressive - I for one am not complainng about a lack of dynamic range with current DSLR's.

[Paul_Gardner]

John Sheehy:
Im sure glad that someone else understands bit depth. I keep telling people to remember the days of the 8 bit monitor and how crappy they looked. I have programmed in assembly for years and while I haven't done any graphics in quite a few years I still have my programming manuals.



One of the difficulties with ADCs is that they are current hogs, and possibly the reason that Canon is not implementing 16 bits is the shortened battery run time. I know the DMR has about half the battery run time as the 5D. How much is attributable to the ADCs is unknown to me.

S/N ratio is just that. It has to do with the noise floor of the sensor. The bottom bit of a 12 bit ADC and a 16 bit ADC has the same analog value. Its all the bits in between that make the difference in the smoothness of the image. Think of a ladder that is 16 feet high, one with 8 two foot rungs and one with 16 one foot rungs. They both start at the same place on the ground.

The chip design either CCD or CMOS is only related to the S/N as both are sensitive only to light intensity levels and not color. Color is derived from the filters placed in front of each photocell which when scanned by the processor which knows which filter is over which cell produces a color pattern.

[John_Richard]

Paul - clearly you understand bit depth and the inadequecies of 8bit (as most people do). However there is a huge leap between 8 and 12 bit but only a small step between 12 and 16bit.

[Paul_Gardner]

John Richard:
Most people may understand the inadequices of 8 bit but without a hardware engineering background not why.
12 bits is = 4096, whereas 16 bits is = 32,768 which is a big difference when looking at the bottom of the bit range where the shadows are.
1/8 of 4096 is 512 bits whereas 1/8 of 32,768 is 4096. Now tell me about available shadow resolution. One will resolve the shadows into 512 Distinct values conpared to 4096 values which is the whole range of 12 bits.

[E_J_Peiker]

8 bit = 16,777,216 discrete colors
12 bit = 68,719,476,736 discrete colors
16 bit = 2.8147 E 14 discrete colors

There is a visible difference between 8 bit and 12 bit on graduations of the same color. I personally can not see a difference between 12 bit an 16 bit although it would allow you to do some really srious manipulation before seeing banding.


[ QUOTE ]
16 bits is = 32,768

[/ QUOTE ]
2 to the 16th power is 65536 not 32768 [img]/images/graemlins/wink.gif[/img]

[Pro Photo Store]

Visit the B&H Pro Photo Store for the best Gear Prices and regular specials!

Brilliant Color for any Digital SLR
White Balance Cap
The Pros Agree on ColorRight
What People Are Saying