8 or 16 bit DNG Files on M8

[kschwarz]

Advertisement (gone after registration)

I was wondering if anyone has information regarding whether or not the DNG files represent 16 or 8 bit color information. Leica promotes the A-to-D converter in the M8 as 16 bit (the usual is 8 to 12 bit), but I understand from certain web reviews that the DNG File is 8 bit. To quote <http://www.luminous-landscape.com/reviews/cameras/leica-m8.shtml>:

 

"Leica has published that the M8 is a sixteen bit device. As soon as the first raw files became available though there were postings online that the files were only 8 bit. To get to the bottom of this I once again turned to Dr. Know. ("It's OK Michael, this won't hurt a bit"). What he told me is that the M8's DNG files are definitely 8 bit, not 16 bit. They may be 16 bit somewhere upstream of the DNG file, but the output file is in 8 bit format. I asked if there were any true 16 bit files from any camera. he said that he had never seen one (and he's seen them all). He said that files from Phase One backs are 14 bit (they too claim to be 16 bit), Canon raw files are 12 bit, and that various other cameras range from 8 to 12 bits."

 

This is a big issue, as an 8 bit file allows for only 256 levels in each channel, whereas 16 bit allows for >65,000. This difference is one of the main reasons for using RAW files in the first place, as JPEG is limited to 8 bit. The file size of the M8 RAW (about 10 MB) suggests that it is indeed an 8-bit file (without compression, one would think that 16 bit files would be about twice the sensor resolution, or about 20 MB). However, compression might be in play or their might be other specifics of RAW that allow for smaller files. If you ask me, this one issue is a bigger show-stopper than the banding and purple cast, as the dynamic range of M8 images will always be limited by the bit-depth of the RAW file. Does anyone have info on this?

[marknorton]

Yes, yes, this has already been covered to death in this forum. Try doing a search.

 

In as nutshell, we think the 12 or 14 bit ADC values are stored as their square root, shifted for greater precision. For example, if there is a 12 bit ADC, that generates values in the range 0 - 4095. Multiply that by 16, add some dither (maybe) and take the square root and you end up with an 8 bit value in the range 0 - 255.

 

We did some analysis which showed the error of companding the data in this way is quite small and there are benefits in terms of storage, processing time, battery life, card capacity and card write times in doing this

 

As for your assertion that the banding, ghost image and purple cast are less significant issues, I beg to differ!

[wattsy]

As for your assertion that the banding, ghost image and purple cast are less significant issues, I beg to differ!

 

Agreed. The old 8 bit v. 16 bit argument seems a distant memory now. My main concern with the M8 is the IR/Magenta cast problem. This is an issue I encountered from the very first shot I took with the camera (I took a snapshot of my wife and attributed the magenta colour of her black suit as a crappy WB or profiling issue). The banding/streaking thing has not been an issue for me at all yet. The ghosting has been an issue (but only in test shots so far). The magenta thing is a potential show-stopper. However, even here I like the camera enough that I will consider a filter solution as a temporary fix (all but one of my lenses are E46 so sticking a filter on the end is not a big deal as an interim solution).