Pages: 1 2 [3] 4 5 ... 8   Go Down

Author Topic: 16 Bit Myth  (Read 52414 times)

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: 16 Bit Myth
« Reply #40 on: December 29, 2011, 04:56:34 pm »

A photon is either present or not. At least according to my rusty physical knowledge.

A hypothetical "photon counting" sensor that had high enough spatial resolution that each sensel in practice either saw 1 or 0 photons (and practically never more than 1) would need no more than a 1-bit ADC to capture every nuance of tonality and DR there is to capture.

When there is noise present (either naturally or artificially), the barrier between spatial resolution and "level resolution" gets blurry. Newspapers and inkjets can deliver more or less high-quality gradations by varying the density of dots that are either "colored" or "noncolored". For this to work, the spatial resolution must be high enough that level information can be "encoded" into high-frequency noise that is either invisible or less annoying.

The wikipedia article on dithering is quite enlightening:
http://en.wikipedia.org/wiki/Dithering
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Re: 16 Bit Myth and the DR of chemical (film) pixels
« Reply #41 on: December 29, 2011, 05:10:45 pm »

A hypothetical "photon counting" sensor that had high enough spatial resolution that each sensel in practice either saw 1 or 0 photons (and practically never more than 1) would need no more than a 1-bit ADC to capture every nuance of tonality and DR there is to capture.

The wikipedia article on dithering is quite enlightening:
http://en.wikipedia.org/wiki/Dithering
Indeed: as I am fond of saying, film used 1-bit pixels, since each silver halide-clump either reacted to light or not, and yet (sometimes with the help of blurring through dye-cloud diffusion), film achieves more DR than "1-bit" suggests. The moral being that with any practically relevant meaning of the words, the dynamic range of a printed image is not at all limited to the "engineering" DR of the individual pixels.

Thanks for the reference on dithering: it gets discussed far, far less around here than resolution or dynamic range, but deserves some attention. Maybe Michael could find someone to write an essay on dithering for this site? Maybe he could invite one or two of our more technically competent forum participants to contribute?
Logged

LKaven

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1060
Re: 16 Bit Myth
« Reply #42 on: December 29, 2011, 05:14:37 pm »

Reminds me of Fossum's proposal for a sub-diffraction layer "gigapixel" sensor with "jots" -- one photon per sensel.  Since he discusses similar issues, I'm linking to the paper:

http://ericfossum.com/Publications/Papers/Gigapixel%20Digital%20Film%20Sensor%20Proposal.pdf

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11288
    • Echophoto
Re: 16 Bit Myth
« Reply #43 on: December 29, 2011, 05:16:08 pm »

Hi,

You are absolutely right! The correct interpretation (in my humble opinion) would be that DR (defined as FWC/readout noise) would exceed 14 bits on a full frame 24 MP Exmoor sensor using todays technology. That may necessitate a fifteenth bit.

Best regards
Erik


There are no fractions of bits (or "hairs" of bits) in any A/D converter. An analog input signal is either truncated (rounded down) or extended (rounded up) to an integer value.

If given the choice of a sensor & A/D converter system which converts its signal into an 8-bit data set or a 14-bit data set extrapolated to 16-bits, I'll take the 14-bit system. No myth there.
Logged
Erik Kaffehr
 

Doug Peterson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4169
    • http://www.doug-peterson.com
Re: 16 Bit Myth
« Reply #44 on: December 29, 2011, 05:21:18 pm »

- The statement is true (the A/D convertor is 16 bit).
- The implied advantage is true (that the file has more tonal smoothness and holds up to post processing better).

The specifics of the entire imaging chain and the usefulness of comparing using a 16 bit vs. 14 bit convertor in any given camera are quite complex and include proprietary and very in-depth knowledge (e.g. are all 14 bit A/D convertors identical in impact on image quality when placed into an imaging chain, are there any additional effects that need to be understood when adding in proprietary dark-frame subtraction technology, do all A/D convertors produce identical amounts of heat and are all available in the same physical sizes/shapes, are competing "14" bit systems using all of their "14" bits fruitfully).

We frequently have very in-depth technical conversations (as full and frank as we have knowledge/experience to offer) with the customers who are very technically minded and interested in such things.

Usually though we simply send raw files or have the client test (in the real world) the camera they are considering so they can make their own conclusions.

Bottom line: if someone buys a digital back because they have been lead by the graphic to expect better tonality and file robustness they have not been misled. So I think for marketing purposes it's a fair distillation of the product features.

However, this thread has convinced me the use of the word "true" was not the right way to go. I've updated the graphic accordingly.

Doug Peterson (e-mail Me)
__________________

Head of Technical Services, Capture Integration
Phase One Partner of the Year
Leaf, Leica, Cambo, Arca Swiss, Canon, Apple, Profoto, Broncolor, Eizo & More

National: 877.217.9870  |  Cell: 740.707.2183
Newsletter | RSS Feed
Buy Capture One 6 at 10% off


P.S. The end of the year is extraordinarily busy for us (tax law in the US means a lot of photographers make equipment purchases at the end of the year) so I will likely not be on LL the next few days. Please don't interpret this as avoidance - I think we (Capture Integration) have shown repeatedly we want to be a transparent and honest dealer.
« Last Edit: December 29, 2011, 05:27:12 pm by dougpetersonci »
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: 16 Bit Myth
« Reply #45 on: December 29, 2011, 05:37:39 pm »

The specifics of the entire imaging chain and the usefulness of comparing using a 16 bit vs. 14 bit convertor in any given camera are quite complex and include proprietary and very in-depth knowledge
Did anyone try to set the 2 lsb to zero, or replace them by random noise prior to raw development? If it has no visible differences from the untouched file when both are developed using identical raw development, that would be an indication that bits#15-16 are of little importance.

Of course, one could carry out measurements using suitable targets for those two cases, or view the difference-file (to see if there are recognizable image features in the difference).

I'd say you can make a pretty good case for or against 16-bit files playing a significant role in any given camera using such methods. Is it worth it, or will the easily fooled be fooled, while the sceptics will remain ... sceptic? I dont know.

-h
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Re: 16 Bit Myth
« Reply #46 on: December 29, 2011, 05:38:42 pm »

- The implied advantage is true (that the file has more tonal smoothness and holds up to post processing better).
More tonal smoothness than using the same sensor (with at most 12-13 stops of DR in the case of all DMF sensors I know of) but with a 14-bit ADC, as is the case in some DMF backs?

Or more than with some 35mm format DSLR's with 14-bit ADC output but reportedly greater per pixel DR, when the images are compared at equal PPI (say with crops to equal pixel count)?

I refer to equal PPI and equal pixel count crops to factor out the legitimate advantage of having more pixels, and so getting equal print size with higher PPI: that is a legitimate IQ advantage, but is nothing to do with recording individual pixel output signals that have at most 13 stop DR using 16 rather 14 bits; it is instead related to the benefits of dithering.
« Last Edit: December 30, 2011, 11:01:35 am by BJL »
Logged

LKaven

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1060
Re: 16 Bit Myth
« Reply #47 on: December 29, 2011, 06:33:56 pm »

- The statement is true (the A/D convertor is 16 bit).
- The implied advantage is true (that the file has more tonal smoothness and holds up to post processing better).

What implied advantage?  There is no implication from a 16 bit A/D converter to more tonal smoothness.  There are many other factors that are responsible for that quality.  

Quote
The specifics of the entire imaging chain and the usefulness of comparing using a 16 bit vs. 14 bit convertor in any given camera are quite complex and include proprietary and very in-depth knowledge (e.g. are all 14 bit A/D convertors identical in impact on image quality when placed into an imaging chain, are there any additional effects that need to be understood when adding in proprietary dark-frame subtraction technology, do all A/D convertors produce identical amounts of heat and are all available in the same physical sizes/shapes, are competing "14" bit systems using all of their "14" bits fruitfully).

With a read error of 2e for the Exmor sensors, not to mention a higher QE, I'd say they are.  Otherwise, I'd expect to see your MF sensors perform nicely at ISO 6400, without Sensor+.  

Quote
We frequently have very in-depth technical conversations (as full and frank as we have knowledge/experience to offer) with the customers who are very technically minded and interested in such things.

Usually though we simply send raw files or have the client test (in the real world) the camera they are considering so they can make their own conclusions.

There are lots of good reasons to buy your products, true reasons.

Quote
Bottom line: if someone buys a digital back because they have been lead by the graphic to expect better tonality and file robustness they have not been misled. So I think for marketing purposes it's a fair distillation of the product features.

I've been led to expect only one thing: that there are 16 bits of beneficial image information to be gained by buying your camera.  Well, is there or isn't there?  

Reductio ad absurdum argument: Why not have 24 bit A/D converters and advertise that?  If people are led by that -- falsely -- to expect better tonality and file robustness, they have surely not been misled?  Either there is a benefit or there isn't, and if you are implying that there is one when there isn't, you are culpable.  

Quote
However, this thread has convinced me the use of the word "true" was not the right way to go. I've updated the graphic accordingly.

True as opposed to what?  

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1882
    • http://www.guillermoluijk.com
Re: 16 Bit Myth
« Reply #48 on: December 29, 2011, 07:53:26 pm »

Did anyone try to set the 2 lsb to zero, or replace them by random noise prior to raw development? If it has no visible differences from the untouched file when both are developed using identical raw development, that would be an indication that bits#15-16 are of little importance.

I have done that with Canon 40D and Pentax K5 RAW files, comparing 12 vs 14 bits in these cases. For the Canon the 2 extra bits were irrelevant, they didn't add anything to image quality because noise is well above a 12-bit interval even at ISO100. For the Pentax K5 the 2 extra bits were indeed useful, avoiding posterization that began to show up when performing 12-bit development.

Since no digital back has lower noise in the deep shadows (read noise) than the K5 at a pixel level, I dare to conclude those extra 2 bits from 16-bit digital backs are just marketing, as were the 14 bits in the Canon 40D. But it's still very cool to have a 16-bit machine!!!

If anyone is interested I can upload a version of DCRAW with a bits decimation option to do personal tests.

Regards
« Last Edit: December 29, 2011, 08:01:42 pm by Guillermo Luijk »
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: 16 Bit Myth
« Reply #49 on: December 29, 2011, 07:58:56 pm »

Since no digital back has lower noise in the deep shadows (read noise) than the K5 at a pixel level, I dare to conclude those extra 2 bits from 16-bit digital backs are just marketing, as were the 14 bits in the Canon 40D.
Did you also try different ISO values to see if in-camera processing of raw was able to benefit from more bits than the sensor appears to make use of in itself?

iso3200 might be a straight "bitshift left" of iso1600, but is it certain that no other iso values are obtained by multiplying the data with values < 1?

-h
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1882
    • http://www.guillermoluijk.com
Re: 16 Bit Myth
« Reply #50 on: December 29, 2011, 08:04:16 pm »

Did you also try different ISO values to see if in-camera processing of raw was able to benefit from more bits than the sensor appears to make use of in itself?

Nope, anyway I don't think there is in-camera RAW processing in recent cameras, at least I never found evidences of it (only the well know Sonys, and Sony quickly allowed the user to disable this NR via firmware).

deejjjaaaa

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1170
Re: 16 Bit Myth
« Reply #51 on: December 30, 2011, 01:37:17 am »


Reductio ad absurdum argument: Why not have 24 bit A/D converters and advertise that?

Pentax K10D (Sony 10mp CCD) once was ad'd as having 22-bit A/D  :)
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: 16 Bit Myth
« Reply #52 on: December 30, 2011, 03:42:05 am »

Nope, anyway I don't think there is in-camera RAW processing in recent cameras, at least I never found evidences of it (only the well know Sonys, and Sony quickly allowed the user to disable this NR via firmware).
Surely, the ISO values of >1600 tend to be obtained by digital multiplication, at least in Canon cameras?

My point was that perhaps also 1/3 stop intermediate ISO values are obtained by multiplying a limited number of analog gain values by 4/3, 2/3 etc?

-h
Logged

Bryan Conner

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 526
    • My Flickr page
Re: 16 Bit Myth
« Reply #53 on: December 30, 2011, 03:45:26 am »

Pentax K10D (Sony 10mp CCD) once was ad'd as having 22-bit A/D  :)

You are correct!  So, does this mean that the modern MF digital backs, that are advertised to be 16 bit, are not capable of producing files of the same (or better) quality as the Pentax K10D from 2006?  What happens in the future when, and if, a fully 16 bit back is introduced?  Are the advertisements going to state "now, we finally really have a full, complete 16 bit image pipeline!".  That will devalue the impression that was put forth in the current advertisements.

I think that it would be smarter to simply state that the image quality speaks for itself.  Make sample raw files downloadable and invite potential customers to evaluate them.  Also, allow customers to use a demo back at the store and take the files home with them. There is no use to play word games/mind games with customers.  Especially with the high quality of Phase One Digital backs.  It reminds me too much of the snake oil/sleazy used car salesman tactics.
Logged

Doug Peterson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4169
    • http://www.doug-peterson.com
Re: 16 Bit Myth
« Reply #54 on: December 30, 2011, 07:47:05 am »

I think that it would be smarter to simply state that the image quality speaks for itself.  Make sample raw files downloadable and invite potential customers to evaluate them.  Also, allow customers to use a demo back at the store and take the files home with them. There is no use to play word games/mind games with customers.  Especially with the high quality of Phase One Digital backs.  It reminds me too much of the snake oil/sleazy used car salesman tactics.

Can I assume that you've never dealt with us (Capture Integration) as a customer before?

Doug Peterson (e-mail Me)
__________________

Head of Technical Services, Capture Integration
Phase One Partner of the Year
Leaf, Leica, Cambo, Arca Swiss, Canon, Apple, Profoto, Broncolor, Eizo & More

National: 877.217.9870  |  Cell: 740.707.2183
Newsletter | RSS Feed
Buy Capture One 6 at 10% off

Doug Peterson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4169
    • http://www.doug-peterson.com
Re: 16 Bit Myth
« Reply #55 on: December 30, 2011, 07:50:47 am »

For clarity this is the advertisement in question:

Doug Peterson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4169
    • http://www.doug-peterson.com
Re: 16 Bit Myth
« Reply #56 on: December 30, 2011, 07:58:02 am »

Again, I'm open for suggestions.

The problem to be solved is explaining that the 22mp files from this camera have better tonal smoothness and better ability to hold up to post processing than the 5DII that 90%+ of our potential clients assume is the same because it also has 22ish megapixels. That must be done in about 15 characters and in a way that 90% of people viewing the ad will understand instantly and without further (immediate) explanation (they can of course always ask for more details when they contact us).

Brainstorming here:
"Large Sensor"
"Flexible Raw File"
"Great Tones"
"Smooth Tones"
"Beautiful Files"

None of those strike me as a great substitution. But if ya'll feel strongly one of them does a better job solving the marketing problem outlined above let me know.

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: 16 Bit Myth
« Reply #57 on: December 30, 2011, 08:02:56 am »

None of those strike me as a great substitution. But if ya'll feel strongly one of them does a better job solving the marketing problem outlined above let me know.

"N stops of Dynamic range*)"?
*)According to DXO "display" measurements
Logged

Doug Peterson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 4169
    • http://www.doug-peterson.com
Re: 16 Bit Myth
« Reply #58 on: December 30, 2011, 08:09:15 am »

"N stops of Dynamic range*)"?
*)According to DXO "display" measurements

A great suggestion but:
- I don't agree that the DXO measurements fully explain the useable range captured (two raw files can have a similar measure but in one the shadows are clumpy, tonally choppy, and inaccurate in color while the other is gaussian/film-like, tonally smooth, and accurate in color - the difference between quantitative engineering measurement and qualitative photographic aesthetics)
- DXO measurements aren't available for all cameras (especially legacy systems like this one)

"High Dynamic Range" maybe, except that photographers now largely associate that phrase with ugly HDR abominations on Flickr.

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11288
    • Echophoto
Re: 16 Bit Myth (P45 vs Pentax K5)
« Reply #59 on: December 30, 2011, 08:39:18 am »

Hi,

Marc McCalmont, one of the frequent posters on this forum, posted raw image from his Pentax K5 and his P45. My understanding is that he felt that the Pentax K5 had better image quality.

I checked those image using Lightroom and this is what I found:
http://www.luminous-landscape.com/forum/index.php?topic=50895.msg421886#msg421886

Best regards
Erik


A great suggestion but:
- I don't agree that the DXO measurements fully explain the useable range captured (two raw files can have a similar measure but in one the shadows are clumpy, tonally choppy, and inaccurate in color while the other is gaussian/film-like, tonally smooth, and accurate in color - the difference between quantitative engineering measurement and qualitative photographic aesthetics)
- DXO measurements aren't available for all cameras (especially legacy systems like this one)

"High Dynamic Range" maybe, except that photographers now largely associate that phrase with ugly HDR abominations on Flickr.
« Last Edit: December 30, 2011, 09:08:47 am by ErikKaffehr »
Logged
Erik Kaffehr
 
Pages: 1 2 [3] 4 5 ... 8   Go Up