bclaff Loc: Sherborn, MA (18mi SW of Boston)

I have a protocol for gathering raw data files that I feed into my analysis software.
I personally gather most of the data but many people have also delivered data sets to me for analysis.
If anyone has access to a camera that is not represented at PhotonsToPhotos they would email or PM for details.

Ysarex Loc: St. Louis

He's off base and we've been through all this before and before. Some helpful reading:
https://www.strollswithmydog.com/dynamic-range-and-bit-depth/
https://www.strollswithmydog.com/how-many-bits-to-fully-encode-my-image/
https://www.strollswithmydog.com/sub-bit-signal/

Read all three. Note that in the final article the author begins with the question: "My camera has a 14-bit ADC. Can it accurately record information lower than 14 stops below full scale?" The answer is yes. He explains why.

selmslie Loc: Fernandina Beach, FL, USA

On closer inspection of the RawDigger plots I posted in https://www.uglyhedgehog.com/t-761295-1.html#13643423 I have found that, although the 12-bit raw values are about ¼ as high as the corresponding 14-bit raw values, they still cover the same range of luminance values within the scene. In other words, a 14-bit raw value of 100 becomes a 12-bit raw value of 2. That ratio carries on through the full DR.

So the 12- and 14-bit raw files actually capture the same range of scene luminance values and therefor capture the same scene DR. That is what Bill Claff pointed out earlier.

There is further confirmation of this in the RawDigger information in that in both raw files claim about the same 2MP to be "UnExp" which means that it is below some threshold non-zero raw value. But that threshold is also 4x greater for the 14-bit file than for the 12-bit file.

So I will retract my earlier comment that the two missing bits represent a loss of recoverable shadow information.

The difference is that by dividing all of the raw values by 4, the 12-bit representation is less precise. Tonality adjustments made during the raw conversion on the computer will be smoother for a 14-bit raw file than for a 12-bit raw file. And recovery of shadow values will be cleaner from the 14-bit file than from the 12-bit file.

This leaves open the question of whether the noise is related to DR or just to the performance of the sensor and the camera's noise suppression software as it converts the analog signal to a digital record.

And how can we reconcile the differences between the DR values presented by DxO and Photons to Photos? Is DxO simply wrong or are they assuming a different noise floor.

Ysarex Loc: St. Louis

Nope:
https://www.strollswithmydog.com/dynamic-range-and-bit-depth/
https://www.strollswithmydog.com/how-many-bits-to-fully-encode-my-image/
https://www.strollswithmydog.com/sub-bit-signal/

selmslie Loc: Fernandina Beach, FL, USA

I can see that in the RawDigger data when I select a small sample. When the SD is high in relation to the average it looks noisy.

That can explain why they might overstate the DR. Most of us are thinking of a slightly larger image at 300 ppi because we can't actually focus on an 8x12 image from 12 inches without reading glasses.

Thanks for all of the feedback. I has been a learning experience for many of us. I will be looking at more of the documentation on your website after a good night's sleep.

selmslie Loc: Fernandina Beach, FL, USA

A 16-bit raw file would not be much help either.

Since many of us capture images rom scenes that don't cover more than 7 or 8 stops, much of this is academic.

There are still very simple ways to suppress the noise in PP as well as to simply cover up the ugly shadows by raising the black point.

But it is interesting to watch the progress of technology by comparing the performance of new cameras to each other and to older models.

TriX Loc: Raleigh, NC

Thank you Bill. I believe I’ve read all the further reading links on the DR chart, and I recommend them to anyone who wants a deeper understanding of this subject. I appreciate you joining the discussion.

I admit that I am still struggling with the idea that the resolution or DR of an ideal 14 bit A/D is 14.5 bits. In fact, with all the A/Ds I’m familiar with, we usually consider it to be n-.5 because of the uncertainty of the LSB, and I cannot find a reference where the resolution is ideally anything other than n. In fact, in actual use, due to the uncertainty of the LSB, quantization error, non linearities and of course shot and read noise, it’s almost always less than the theoretical value, hence my lack of understanding how a 14 bit A/D can produce a DR in excess of 14 bits, especially in the real world as opposed to the ideal. Can you comment and help me understand or provide a reference please?

Strodav Loc: Houston, Tx

I agree each individual color channel can only have a DR of 2^N, but the DR we are talking about is a single number generated from 3, red, green and blue 2^N channels. So assuming one DR number from a triplet where each is 2^N can only equal 2^N is simplistic. If they do not align perfectly in highlight and black given a standards light source with the desire to achieve a certain color balance, which is unique to each camera brand, and the desire to avoid colored or speckled noise, they may not align perfectly. So you can have, say blue saturate before green and green saturate before red. Same at the shadow end.

TriX Loc: Raleigh, NC

Further to my last past. During the 12 years I spent at Tektronix, we didn’t test the DR of cameras, but since Analog Devices (a major manufacturer of A/Ds,) was my customer, I’ve had occasion to test hundreds of A/Ds, and in fact we developed a new testing system when A/Ds and D/As exceeded the DR of conventional instrumentation We never saw an A/D with a DR as good as its theoretical resolution in actual practice (1 bit = 1 stop = 6 db), much less better. Hence the reason I trust the DR data of photons to photos and cannot understand how DXO produces 15.5 stops DR from 14 bit data. I’ve re read Bills explanation of the difference in testing methodology (6 times), but I STILL don’t get it.

selmslie Loc: Fernandina Beach, FL, USA

That should be “... 12-bit raw value of 25.” Sorry for the typo.

Architect1776 Loc: In my mind

selmslie Loc: Fernandina Beach, FL, USA

A closer look at the RawDigger plots clarifies that "UnExp" actually does mean pixels with 0 values.

There are 1M unexposed blue pixels in both the 12-bit and 14-bit capture. The other three channels combined show another 1M unexposed. There are also a tiny number of pixels "OvExp" in all 4 channels. That's not surprising since there simply isn't much blue in the image and it is lit by incandescent light. The blue isn't underexposed. There just isn't much of it.

The plots are set to show 15 stops (+4 to -11) for the 14-bit capture. The highlights are two stops to the left for the 12-bit version.

What is actually happening at the shadow end is difficult to discern because from -11 to -4 (raw values of about 0 through 64) there are spikes rather than a solid plot which shows from about -4 through the upper limit. This means that the really good part of the raw capture (decent tonality) covers about 8 stops for the 14-bit file but only 6 stops for the 12-bit file.

So even though the levels where noise becomes visible is about the same, the 14-bit file still has a significant edge when it comes to recovering smooth tonality from the shadows.

The rescaled plots make this easier to see.

12-bit

(Download)

14-bit

(Download)

dpullum Loc: Tampa Florida

Selmslie, When testing choice of subject matter is important and when dealing with HDR the camera will make adjustments of exposure when you set the camera menu to HDR.

I suggest you have "DGK Color Tools High Resolution 8.5x11" Chrome SD Professional Lens Test Chart" 3 for $15. or pick a a SIMPLE UNCOMPLICATED subject with light and shadow areas...

Noise: If you took the same photo at same conditions multi times and blended them the "random" noise would cancel out ... averaging out.

And.... Do not forget the new AI de-noise programs which work magic making many noise questions historic and moot... abstract or purely academic.

photon-collector Loc: Tampa Bay Area, Florida

It's all very simple, really. You push the first valve down....the music goes round and round...me or my oh...and it comes out here.

selmslie Loc: Fernandina Beach, FL, USA

That's a good tool for testing resolution and color.

But no single printed chart under uniform lighting will actually show more than about 5 stops of DR. When I was testing film with a Macbeth Color Checker I needed three bracketed exposures to capture a 12-stop range of tones.