B&W Film Dynamic Range

[Neal Shields]

I think the big problem of applying Moore's law to camera sensors is that in a memory chip or a CPU if you have a dead transistor you simply "map it out". On a sensor you can only have so many dead pixels before the entire sensor is junk.

The bigger the sensor and the more pixels the more likely that you will have to throw an unacceptable number away.

With some of the sensor technologies, information is read out by passing the information down the row from one pixel to another, like people filing out of a movie theatre. If you have a dead pixel, all pixels upstream can't communicate. I don't think CMOS has this problem but it comes at the cost of having more non light sensetive areas on the sensor. I.E. if you don't pass from one pixel to another you have to have highways to move the data. I think I remember that there is a problem with passing the electrons to lower layers of the chip but I don't remember why.

[jetcode]

On the 5D it seems that the sensor has a non-linear response around this area (steeper curve?).

You are dead on in assessing that all analog sensors exhibit some form of linear and non-linear transfer curves. Much of sensor conditioning pertains to making corrections in gain, linearity, and precision. Every sensing system whether analog or digital conforms to nature of analog systems. Beyond 18 bits of resolution is the noise floor or the random chaotic noise that permeates the universe. 24 bit converters are highly filtered to achieve higher resolutions. Sensors are notorious for non-linearity. A modern transistor has a very narrow range of linearity but when used correctly can produce a beautiful linear function. Half the battle in engineering is working with inherent limitations in nature and device physics.

Joe

[jetcode]

I think the big problem of applying Moore's law to camera sensors is that in a memory chip or a CPU if you have a dead transistor you simply "map it out". On a sensor you can only have so many dead pixels before the entire sensor is junk.

The bigger the sensor and the more pixels the more likely that you will have to throw an unacceptable number away.

With some of the sensor technologies, information is read out by passing the information down the row from one pixel to another, like people filing out of a movie theatre. If you have a dead pixel, all pixels upstream can't communicate. I don't think CMOS has this problem but it comes at the cost of having more non light sensetive areas on the sensor. I.E. if you don't pass from one pixel to another you have to have highways to move the data. I think I remember that there is a problem with passing the electrons to lower layers of the chip but I don't remember why.

There are no bad transistors in CPU's, there are failures in logic and that is why no one can find the BIOS engineer at Intel. Mapping out bad memory happens in high density flash, there is no mapping at the CPU level or in RAM, every cell has to work or the chip is tossed. There are flash components that are 100% functional and are very expensive and used for mission critical applications.

bucket brigade devices (chained, movie theater) are noisy and no longer employed as the noise floor is amplified by each cell as the signal passes from one cell to the next. These devices were used for analog delay. CCD's and CMOS are not bucket brigade, they are analog sensors cells operated in parallel limiting the over all noise floor to a single cell. I think you are referring to the multiplexing that is necessary to interface huge arrays of pixels in sensors. It's the same strategy that is used to interface huge arrays of memory. Careful design will limit the degradation of an analog signal in a multiplexed interface.

[marschp]

Sorry Paul. ....

I would say that if you are only acheiving 5 stops, the problem is inherent in your exposure and Raw processing technique and not in the capture devices.

Continuing to state 5 stops is nothing more than nonsense and is just plain incorrect. Sorry.

David - I checked those links - very interesting, but I'm still not convinced. It feels to me a little like when my Lexus dealer tells me that the RX400H can do 34.5mpg - that might be right in the lab, but my experience tells me its not going to happen in real life (I get 27mpg).

I just took two sets of (real life) test shots to challenge myself, and compare with the DPR claim. The scene metered a total dynamic range of 7.6 stops. The Jpeg, on a generous assessment, seems to have captured 6 stops before maxing out. The overall image appears dark and the distinction among the shadows is poor. I have bracketed jpegs of the same scene that look better tonally, but the sky is blown out and not recoverable.

The RAW version did better. Using the ACR settings suggested in the DPR review for the 5D I was able to produce a histogram that encompassed the full 7.6 stops range of the image as metered. However, this produced an image that I would class as unusable. The DPR test suggested settings seem to sacrifice mid-tone luminosity and raise the shadow detail (and noise with it) in order to maximise DR - the result is a flat dull image that lacks contrast in the shadows and the highlights. (By the way, I was pretty surprised to see that the ACR default settings used in the DPR test produced a RAW DR that was about 1 1/2 stops WORSE than the jpeg test).

I'm intrigued by the DR response curves that DPR produce - particularly the non-linearity they all display. I'm by no means an expert in this field (as you probably can tell), but maybe its that non-linearity in the shadows and (to a lesser degree) in the highlights that explains the difference between 'theoretical' DR (the 10-11 stops that DPR state) and 'usable' DR (the 5 and a bit stops that I'm talking about).

Next trip out with my 5D, I think I'll still be taking my grad filters.

Cheers

Paul

[David Luttmann]

Paul,

When you start with an image with high DR, you have an inherently low contrast image to start. You then have the ability to adjust it afterwards to a contrast level that you find suitable. As to noise in shadows, there is far less noise in those shadows than there is with film grain.

That said, you just did the test yourself an obtained 7.6 stops with a 12 bit system. A moment ago it was only 5 stops. Amazing how these figures change when subjected to scrutiny ;-) I find better results with Capture One to the tune of about .5 stops.

That said, I've had no problem achieving 8 stops or more which is more than I've ever obtained from ANY Fuji or Kodak transparency film.....and definitely more than 5 stops!

[marschp]

Paul,

That said, you just did the test yourself an obtained 7.6 stops with a 12 bit system. A moment ago it was only 5 stops. Amazing how these figures change when subjected to scrutiny ;-)

Yes, but my point is that the 7.6 stops image is unusable, no matter what post processing treatment. So I have to conclude that the 7.6 stops of DR that I thought I had, is in reality not letting me produce a good image. On the other hand, if I had approached the shot as though my sensor could only manage 5 and a bit stops, then I would have achieved a much more satisfying image, either through the use of grads, or through the blending of several images via HDR or layering.

Here's a thought: my kind of landscape photography takes place in low light conditions, but I'm still trying to capture a broad range of light - the unfiltered histogram would no doubt have a hump at either end of the range, with not much in the middle. Generally, I would apply 2 to 5 stops of ND grad to bring the scene within what I regard as the 'useful' DR of my sensor - i.e. 5+ stops. So lets say the typical scene I'm shooting is 7 to 10 stops - supposedly within the 5D DR claimed by DPReview. Yet, the bulk of the contrast is at the extreme ends of the DR - just where the sensor becomes the most non-linear. Doesn't that suggest that the sensor is least useful just where I need it most - i.e in the area of highlight contrast and shadow contrast, and that's why I'm struggling to accept the 'usability' of the full DR of the sensor.

Paul

[tim atherton]

Yes, but my point is that the 7.6 stops image is unusable, no matter what post processing treatment.
Paul

the 20 stops of TMX isn't very usable either (unless you are photographing things like nuclear explosions)

The point is that it's there if you need it.

With B&W film you chose you range/development depending on the subject matter and the look you want - sometimes you compress it, sometimes you expand it, sometimes you just go for the middle of the road

[paulr]

This thread is reminding me of how cool b+w film is. I have no particular attachment to color film, but it seems like a satisfying substitute for b+w is long, long way off.

[Joanna Carter]

Yes, although I have taken some stunning shots in colour, I have had to work hard to capture the full DR that was present; especially where the highlight/shadow was not filterable unless I had a triangular or oval shaped grad filter :-) On occasions like that I have resorted to taking more than one trannie and compositing the scans into one "HDR" image.

By contrast, I have also had the immense pleasure of taking a B&W shot with some 11 stops of DR, developing it at N-2, and playing with split contrast printing techniques to produce the most wonderful of images.

My problem? My eye tends to discover more colour pictures than B&W but the B&W, when I find them are, by far, the most satisfying.

[Helen Bach]

Yes, although I have taken some stunning shots in colour, I have had to work hard to capture the full DR that was present; especially where the highlight/shadow was not filterable unless I had a triangular or oval shaped grad filter :-) On occasions like that I have resorted to taking more than one trannie and compositing the scans into one "HDR" image....

Slightly off topic, but why not shoot colour negative if you want a good dynamic range?

Best,
Helen