Bit-depth, noise and dynamic range in CMOS sensors

[rosuna]

Advertisement (gone after registration)

This article is very interesting:

 

Noise, Dynamic Range and Bit Depth in Digital SLRs

 

Martinec cannot find any image quality difference between 14-bits and 12-bits CMOS cameras, mostly due to noise.

 

Emil Martinec bio:

 

Emil Martinec

 

.

[Artichoke]

which is more technical than I can fully comprehend, but which confirms a long standing prejudice that I have had regarding CMOS sensors vs CCD sensors

the digital look that I see with CMOS files stems from the extra processing needed to deal with their inherent higher noise levels

increasing bit depth does not offer much of an advantage for CMOS sensors

Leica was wise in going with Kodak's CCD for the M8 and they did a remarkable job with their DNG fabrication

[rosuna]

It is interesting to read the comments of Martinec about tonal compression.

 

I don't know if Martinec analysis applies to CCD sensors, but the 14-bits A/D of the M8 seems to be pretty good. Then, the tonal compression is a great idea. It is not a "lossy" treatment of the original data! The problem with the M8 is the excessive compression adopted: 8-bits files include only 256 levels. Nikon sheme is based on a 10-bit compression.

 

Software noise reduction is problematic, but there are a lot of "hardware" noise handling tricks on play in CMOS sensors.

[krabat]

I still do not understand why Leica does not offer at least an option (for those who want it) to switch to full-resolution storage of the image data. I know, a lot of people claim that the difference would not be visible, and that fabulous images can be created using the compressed 8-bit data, however, I would feel more comfortable with the opportunity of full 12 or 14 bits. Even the RAW files of the good old fashioned Digilux 2 are not compressed to 8 bits!

 

Cheers, Peter.