« It's 9/11... | Main | Turnley at the Times »

Wednesday, 12 September 2012

Comments

Ctein:
Is there a rule of thumb to go from EV to photoelectrons per second?

(So I can translate your photoelectron bins to a real camera/sensor.)

Mildly off topic, but I was wondering... would it be possible to produce a sensor that was able of working like a regular color sensor, as well as a truly monochromatic sensor? I don't know enough about how sensors work so that's why I am asking. I don't know if that's something that would be physically possible. If a pixel site, or photo site, or whatever it's called, is predetermined on the sensor to read red, is that due to the physical nature of the the sensor, or is the fact that that pixel reads red something that could be dictated through software? I am going to assume that it can't, but I might as well ask

Or say one had a sensor like the Foveon sensor, where each photo site reads red, green, and blue, unlike the bayer-array.. As I understand it, this is because unlike the bayer type sensor, where each pixel reads either red, green, or blue separately and then the proper color is determined through software, the Foveon sensor is built kind of like a layered cake. So when light hits the sensor, each pixel's color is determined by reading RGB info at every pixel, as light has to pass through a red sensor layer, and a green and blue at each site. It does not work by interpolating color information like the Bayer-array sensors do, but produces true RGB info at each pixel site.

If that is the case, could one theoretically make a sensor that was similar to this, but that also had a layer that simply read luminance as well? Or maybe the sensor would use the lab color space instead of RGB, so each pixel had a layer for channel a, channel b, and then the Luminance channel. If one then wanted to make a b&w image, the A and B channels could then be turned off and sensor would then read only luminance, essentially making it a dual color and monochromatic sensor, where setting the camera to b&w would actually be making it a true monochromatic camera.

Does this even make sense? I am just kinda talking out of my rear end here as this thought literally just popped into my head after reading this article. I am most likely using a few terms improperly as well, so I this makes no sense whatsoever do not feel bad by telling me that I'm just nuts.

It just seemed like an interesting idea, so I figured I throw it out there... Any thoughts?

How about the referring to non-digital imaging as "analogue?" Does that get your dander up -- as it does mine?

"As a thought experiment, let's imagine an ideal, noiseless monochrome sensor that converts each incoming photon to a photoelectron, and where each pixel can hold 1000 photoelectrons. Then its dynamic range is 1000:1 because the smallest signal it can record is one electron and the largest is 1000."

Picking a nit here, isn't the smallest signal zero electrons?

I've got the donkey, I've got the lance, and I'm riding with both. For truth, justice, and Dulcinea!

It bugs me when people think Don Quixote rode a donkey. He didn't, he rode a horse. It probably shouldn't bug me, but it does.

Okay, but do you prefer red wine or white with salmon?

See, this is why I reckon film is a lot simpler!
With film, all I have to worry about is density range!
:p
(walking out, shaking head and yelling "I am not an animal!")...

Thanks, Ctein, for referring the Martinec article and for your "introduction." Beautiful indeed.

After a quick read, I know now how "clipping" got it's name. It's the histogram at either end of the scale being clipped at the shoulder if not right down the middle. That the histogram display is a series of histograms strung together.

Second, that ISO is an amplifier. Its no use cranking up the ISO if it introduces just as much, if not more noise than the signal gain. (Like cranking up the volume of a middling amp to its clipping point. Your fine speakers will faithfully reproduce both sound and noise.) Digital imaging and audio has so much in common. White noise looks just like it sounds. Beautiful illustrations (yours & Emile's). As for ETTR, I don't like cranking up the volume on my amplifier beyond the midpoint (0 db; ISO 800 on my camera).

I know now where social scientists get their physics envy. Your scattergrams are so "well-behaved."

Did I get the first two right, or do I have to reread the article again? I'm absolutely sure about the third.

Actually, Ctien, there is a simple answer. It is: "Modern digital cameras are so good, just stop obsessing and take some &*$%@ pictures, already." :-)

I admit it; I have exposed to the right! Also to the left, depending on the circumstances. Correct exposure is subjective, and its determination is part of what the photographer is supposed to do, both in exposing a photograph and in processing it. We do indeed need rigorous analysis of sensors, especially at the pre-design and design stage of the camera. But exposure decisions, the best ones anyway, are subjective. So, as much as I appreciate your analyses and the objective, observational aspects of science, how does this fact of the event of exposure blend in with exposure and dynamic range, other than to set limits? I will be happy to get the answer in your column next week, and happy as well to read it whether the answer is there or not. The ETTR crowd (and I use that term with purpose) have a lot to answer for in our age of education by Internet.

I had "ETTR" questions but with a quick Google noticed they've already been answered by yourself and Mike.

On the topic of exposure, I would have thought the E-M5's "Highlight & Shadow" mode was a lot more useful than a histogram but I haven't seen much discussion about it. Perhaps when/if you or Mike review the camera?

only 2 comments; musta bin too deep

I'm sorry ctein but as I understand your definitions - and it's quite possible that I haven't(!) - exposure range would be an example of a type of dynamic range; and if it is a type of dynamic range, is it so misleading to use that term?

Don't know if you want to segue into a comparison of film vs digital, but I was pleasantly surprised by my F90x and some expired Fuji 160s print film last weekend. I am certain my D700 would not be able to capture this scene with the same verisimilitude. How many stops was the film able to capture here- 14?

http://www.flickr.com/photos/pgherbster/7964409208/sizes/o/in/photostream/

Thanks for this interesting note. My attention as a pragmatist was most firmly caught by your phrase "...this further trashes the arguments of the "expose-to-the-right" crowd."

Can you expand on this point, and explain in what way the ETTR argument is trashed? And... what *is* the preferred way to determine exposure on a digital sensor? From personal observation I would say my camera is implementing ETTR when set to matrix/evaluative metering. I have been relinquishing control to it as I thought ETTR was optimal for digital sensors -- even though it usually needs post processing to get a realistic look.

Thanks for an article that's interesting, right on point and technically correct. I'm looking forward to your discussion of photon statistics.

"It bugs me when people think Don Quixote rode a donkey. He didn't, he rode a horse."

Maybe he rode a mule. (For those who don't know, a mule is the offspring of a horse and a donkey.)

This is why I love to read Ctein posts: Useful information presented with supreme technical accuracy... combined with the romantic impulse to fight hopeless battles! (As in "a pixel is not a little square")

...He didn't, he rode a horse...

Right. His (the horse's) name was Rocinante. He wasn't a stallion, though. [R]osinante (lowercase) has entered the English language meaning: "a broken-down horse"; "nag" (Webster's Third New International Dictionary, 3 vols.).
Sancho Panza ("his commonsensical peasant squire") rode a donkey, if I remember correctly.

Whatever you choose to call it, the majority of users still don't know how to make use of the knowledge of exposure range, or more importantly, the behavior of the photons between the extremes, i.e. the linearity (or not) of the sensor and how that relates to the non-linearity of the human eye/brain with which we ultimately perceive the image (ohh, I'm getting over my head here I think). Consider the huge volume of "overcooked" "HDR" images on the web, let alone printed or displayed anywhere else (just what is the Dmax of that paper, or contrast ratio of that display? Assuming you've calibrated it. How do you calibrate an iPad display anyway? oh my brain hurts!). And don't get me started on the static exposure range (sensitivity?) of the human eye, which is supposedly 6.5 stops.

For most "image capture device users", the more that the exposure range of your capture device exceeds the range of your display medium, the greater the likelihood of effing up the result.

(Personally, I still print on matte paper, which has a much narrower range than my five year old digital camera, and the final steps in the development of the image, mainly proofing, presents the greatest challenge for me.)

Dear Folks,

I really have to curb my inclination towards throwaway remarks; it gets me into trouble (and good luck with that, he said to himself). This is not a column about ETTR. I did that already:

http://theonlinephotographer.typepad.com/the_online_photographer/2011/10/expose-to-the-right-is-a-bunch-of-bull.html

Emil's article throws a further monkey wrench into it, because he demonstrates that the relationship between data and brightness level is NOT the simple thing the ETTR people think it is.

It was just a quip.

Go read my previous column. I am not going to discuss ETTR any further. A waste of my time (just like ETTR).

~~~~~~~

Dear Bear,

I don't think you entirely understand my column. In fact, no, exposure range is not a "kind of dynamic range" in sensor parlance. But, even if I accepted your premise, you'd still have a situation where "exposure range" refers to exactly what we care about and "dynamic range" could mean several things and most commonly doesn't refer to what we care about. So, bad usage.

~~~~~~~

Dear Bill M.,

"Analog photography" doesn't bother me because it's a familiar construction in English. It's called a reverse neologism. It happens when a new meaning for a word overtakes its old one. For example, originally you had "computers" and "electronic computers." After the latter become dominant, folks dropped the "electronic" and started referring the the original computers as "human computers." Similarly, you may remember when we had "(tele)phones" and "touch tone phones". That become "dial phones" and "phones."

In the same way, as the norm for photography becomes digital, it becomes a useful convenience to make it clear what kind of photography is involved-- silver or silicon. In not too many years, it will automatically be assumed that "photography" means "digital photography" and you'll need the reverse neologism for clarity.

It's not a passing of judgement, it's just language functionality.

pax / Ctein
==========================================
-- Ctein's Online Gallery http://ctein.com
-- Digital Restorations http://photo-repair.com
==========================================

"It bugs me when people think Don Quixote rode a donkey. He didn't, he rode a horse."

Possibly, but we were talking about don Ctein de la Daly City. And don Ctein, he rides a donkey. Named "Roentgenogram."

Mike


Dear Keith B,

One of TOP's readers, Nick Condon, wrote an article related to that:

http://imagerelay.blogspot.com/2011/03/how-perfect-is-perfect-sensor.html

Afraid I haven't reduced it to a simple rule of thumb, though it shouldn't be TOO hard to figure out.

In practice, remember that current camera sensors aren't anywhere near 100% efficient, so one photon doesn't equal one photoelectron.

~~~~~~~~

Dear David D.,

No, because there's no way to distiguish a "zero signal" from no signal at all (as in, for example, the chip isn't even plugged in). So, no data is conveyed by zero photoelectrons; the theoretical minimum is 1.

~~~~~~~~

Dear Noons,

Yup, if you're a mere duffer who doesn't give a damn about any other part of film usage (like contrast, tonality, etc.).

No photographer really needs to understand what's in this or the next column. Not for making photos. But if I simply shouted, "Stop misusing 'dynamic range!'" with no explanation, how many people would pay any attention?

~~~~~~~~

Dear Alan,

Oh, hear, hear! Go, brother, testify!

Unfortunately, there are some windmills to big for even my mighty lance and steed.

Hell, I can't even get people to stop reporting results to three significant figures when they should only be reported to 1.

I just sob softly into my physicist's pillow at night.

pax / Ctein
==========================================
-- Ctein's Online Gallery http://ctein.com
-- Digital Restorations http://photo-repair.com
==========================================

Dear David,

Oh yeah, I wrote a column about that problem of squeezing a pint into a 12 ounce can.

http://theonlinephotographer.typepad.com/the_online_photographer/2009/04/fitting-a-pint-in-a-12ounce-can.html

pax / Ctein

Germans, being what they are, adapted "Dynamic Range" to their photographic vocabulary, abbreviated to "DR".

But we have this beautiful word "Kontrastumfang" - you might translate it with "range of contrast" - and that is what I am still using.

If only to stun those who prefer to use Denglisch instead of their own language when there is no need to.

What's this nonsense about "60 power dB"? It's just 60 dB, period!

Anyways, on the topic of (beneficial) noise, don't know if it is true, but I love the story I once heard about the coining of the term dithering, not that I can tell it as well as whoever told it to me. Seems RAF engineers discovered some instrumentation on their planes was faulty: the gears that drove the panel indicators were corroded and stuck together. But what they found most surprising was that the pilots hadn't informed of the situation. So an investigation was promptly initiated to find out why those irresponsible folks had been flying without watching those instruments. Protocols were revised. Interrogations ensued. The pilots insisted they had noticed nothing wrong.
Long story short, turns out the vibrating plane engines, once on the air, introduced just enough low-level noise to the system to make everything work perfectly well.

Ctein,

It's good to know that someone has the brains to understand the theories that justify my opinions ;)

if 98% of people are using a term in a particular way then that must be correct usage and dictionaries will be revised in due course.

Take away message!

Yada, yada, yada... "multiply the dynamic range by 0.95 to get the exposure range." -Ctein

Oh good, something that I had never even noticed will now bother me forever. You have just stuck a permanent stone in my shoe. Thanks a lot.

In 1989 I briefly had a Canon RC-250 xapshot which was an electronic non digital still camera that recorded to an analog magnetic disk and had a lead-acid battery.

After 2 days I realized that a video camera and a targa board were vastly more practical

THAT was an analogue camera.

"It's just 60 dB, period!" You must be joking

Originally a decibel (dB) was one tenth of a bel (B), i.e., 1B = 10dB where a bel is the attenuation of the signal over a mile of twisted pair telephone cable named after Bell Labs who came up with the idea. Then there were a whole bunch of contortions to make it seem less arbitrary and to work for things other than audio. All I can remember it that you measure signals one way and power another way and that there are two different ways to calculate DB ,field and power, and it's all 10 log except for when it's 20 log.

Somehow 2 ended up being equal to 3 and I stayed an art major.

I just took a peek at this
http://en.wikipedia.org/wiki/Decibel#Video_and_digital_imaging
and recalled the horror of being shown a circuit of a bunch of voltage controlled oscillators and filters and having to plot the signal. In an art school final.

If Ctein is "The Man of LaMancha"..does that make you, Mike, Sancho Panza...or possibly Cervantes (as Miguel is Spanish for Michael)? Better that you are Cervantes than Sancho Panza as the meaning of his two names are not the most flattering :-))

I can not wait until Part II of the dr/db/exp epic. Will we then know the truth? Is bigger better? I am referring to pixel count of course!

Finally, horse, donkey, mule, burro, jackass, jenny, or unicorn, no matter what your ride, it is your mount of choice and ALWAYS the best. Just like the age old discussion of which camera is the thoroughbred of the specie.

Ctein, again you have inspired me to further investigate a topic I did not know I was even mildly concerned about!

THANK YOU, Ctein. This is a pet peeve of mine, and until now I had nowhere to send people when I insisted that "dynamic range" meant something different than "exposure latitude."

If you compare a 5D Mark III with Tri-X, you'll find that the 5D has much less noise, but the film has has way, way more latitude. Put another way, the 5D has more dynamic range, but far LESS exposure range.

Don't even get me started on bitdepth, which is related to dynamic range but NOT exposure latitude. You can represent anything within 1 bit. For example, an offset plate for a photo book is 1 bit. Even light itself is 1 bit—there aren't brighter and darker photons. So when people tell you it's necessary to do everything at 16 bit in Photoshop, you should be instantly skeptical. If you can see any noise in your image (and when can't you?), then by definition your photo has less than 8 effective bits.


Ed. Note: I appreciate the comments so far, but we are not going to get into a debate about power vs. field decibels and the history of the Bel. Not the place. Just letting y'all know.

Mike

To make explicit another way of determining at exposure range (implied in Cteins above though I'm not sure if he disclaims it in a comment) is used by DxOmark.

It is no different from the usage traditionally used in RF systems. In RF systems there are multiple "dynamic ranges" (intermodulation dynamic range, blocking dynamic range, etc). In those cases you just define how you measure the upper limit. Similarly for any given modulation type you define the sensitivity in terms of minimum signal to noise ratio required rather than the MDS (minimum detectable signal i.e. the sensitivity at 0dB SNR) for the system. You specifiy what is needed and the rest of the system parameters flows from that.

If you want to do these calculations in your head. 6dB is one stop and 6dB is one added bit in your "photoelectron count" (a nice symmetry). So you can do the calculation in dB by addition and subtraction.

If one defines the minimum acceptable signal to noise ratio** for a "usable black" in the shadows for your then you can calculate the exposure range from the dynamic range. For example, 20dB is used by DxOMark and seems reasonable in practice.

exposure range = dynamic range - minimum acceptable signal to noise ratio

If 20dB SNR is acceptable for you the exposure range is 2.5 stops less than the dynamic range. This is useful to keep in mind when looking at DxOMark numbers, graphs and examples of different SNR ratios.

** More correctly signal + noise to noise ratio as that's what you measure.

Is there a rule of thumb to go from EV to photoelectrons per second?

No because EV is purely a function of shutter speed and aperture and doesn't take sensitivity into account.

You mean LV (or EV at ISO 100 would be the same figure).

You will be needing this at the end of my quoted line!!

Excellent article. Kudos!

And don Ctein, he rides a donkey. Named "Roentgenogram."

Ha ha ha, good one, Mike.

Dear Michael H.,

Oh, you're an evil, evil man.

I've been ego-scanning for online references to me since 1990. You'd be amazed at the stuff that gets incorrectly attributed to me. Well, no you wouldn't. I became resigned to it eons ago.

I'm especially impressed when heated arguments break out as to what I did or didn't say or mean and, 99% of the time, no one bothers to email me for my opinion. It's not like I'm hard to find...

I know, why ruin a good InterWeb argument with primary sources?

~~~~~~~~

Dear Ben S.,

Bruce Fraser and I discussed this once and we decided that the best way to explain this to people was to think of it thusly: the range (whether dynamic, exposure, input, output, whatever) is the height of the staircase. The bit-depth is how many stairs there are in the staircase.

And even that's not fully correct, as my previous column on noise being your friend (and next week's column) explain. When noise enters the picture, you can usefully record a different number of stairs than bit-depth would predict -- sometimes many less, sometimes many more.

~~~~~~~~

Dear Kevin P.,

Right-o! This is a good example of what I meant about the tricky part of nailing down the shadow end of the exposure range. As it happens, DxO's endpoint seems to match my field tests pretty nicely (we come up with the same exposures ranges to within about a half stop), but that's just me. Someone else might decide 10 dB separation was plenty; someone else who is a low-noise fanatic might insist on 30.

Film speed determinations have similar problems. The standard shadow-speed point of 0.1 d.u. above base+fog is just as arbitrary a definition as 20 dB SNR. Many people arrived at very different personal E.I.'s based on their shadow detail preferences.


pax / Ctein
==========================================
-- Ctein's Online Gallery http://ctein.com
-- Digital Restorations http://photo-repair.com
==========================================

@ Dulcinea's lover

Not to beat a dead horse...

To be a "sancho panza" is not really derogatory since it connotes being down-to-earth, a foil for the "high-blown" and the "quixotic."

As for MC-J de Waukesha being a Miguel Cervantes de Saavedra, you got something there. Don Quixote went "viral" among the literates in Europe when it (Part I) was first published in 1605. We all have read or know of Mike's viral TOP (I & II) satire blog posts.

Ctein, otoh, is more of a Jorge Borges, imo. Both being erudites in their respective fields (physicist, bibliophile) and avocations (printer, novelist).

Ctein, stop doing this.....I spend over a week trying to teach somebody on another forum that a sphere (as in a VR sphere) like this one is a 3D object.....I failed.....so I've packed my horse and rode into the distance, since the world is dominated by the stupid these days. Yeps Ctein the earth's B-ark would be verry, verry big...indeed.

Ed

Sarge,
Hmm, Ctein as Borges. I'm going to have to mull that one over.

Mike

Dear Sarge,

I humbly accept (g).

(And thereby leaving people wondering if Ctein is really entirely clear on the meaning of the word, "humble.")

pax / Ctein

About DxOMark's results and scores, what is published (and widely commented) is their "dynamic range" based on SNR=1 minimum black. I did not see scores where they use an exposure range with a 20dB minimum black ? (just an interview)

Also, looking at DxO measures, conversion from 0dB DR to 20dB ER depends on the camera : if read noise dominates -> -3,3 EV, if photon noise dominates -> -6,6 EV. E.g. present Sony sensors need around -1 EV more than Canon or medium format, because they have less read noise.

Just to illustrate your point that there is no simple conversion...
If only your article could be widely read ! Thanks !

Fred

Whew! {&vbg}

In my comment I wrote "novelist" which was an egregious mistake. Borges didn't write novels but "shorter fiction" ranging from short story to monograph length. The anthology I have—A Personal Anthology by Jorge Luis Borges—being compiled by Borges himself and not chronologically arranged, read like a novel from cover to cover the first time I did.

The comments to this entry are closed.