5

The concept behind the dynamic range definition given by ISO 15739 is based on the ratio of the maximum luminance that receives a unique coded representation （the "saturation" luminance) to the lowest luminance for which the signal to noise ratio (SNR）is at least 1.0.

Can someone help me understand the meaning of the part in bold?

mattdm
  • 143,140
  • 52
  • 417
  • 741
petalse
  • 131
  • 7

2 Answers2

3

The maximum saturation point of a pixel's photodiode in a digital sensor is the point at which it can no longer generate new electrons for additional photon strikes. At an electronic level, the saturation point varies with each type of sensor, as the maximum electron capacity of a given photodiode is dependent upon the physical area of the diode, its efficiency, its tendency to leak electrons, etc. Dynamic range can be described in a variety of ways, however a common way to describe it (often used by camera review sites like DXOMark) is as the ratio between maximum saturation, and the standard deviation of the sensors electronic noise floor.

The ISO 15739 specification you are referring to is describing a specific, standardized approach to describing dynamic range. From what I can find, this standard is a little convoluted, and used terms that don't seem to have an adequate explanation of why they are derived the way they are derived. The saturation point they are referring to, if I understand correctly, is actually 100/140th (71%) of the sensors actual maximum saturation level (as described above in my answer.) The full ISO 15739 specification is not freely available as far as I can tell, so the information I have to go off of is simply derivative works based on the ISO 15739 standard.

If my understanding is correct, however, dynamic range in this standard is described as the ratio between the point where shadow SNR is 1.0 and 71% of the sensors maximum saturation point (where no more electrons can be held in the photodiodes of each pixel.) The saturation luminance is that 71% of maximum saturation. In more detailed terms, ISO 15739 DR is the ratio between a luminance level that matches the noise ratio (say 8 electrons worth) and the 71% of the maximum saturation point, which for say Canon's 18mp sensors would be about 22000 electrons * .71, or 15620 electrons. That is a ratio of 1952.5:1 in terms of electrons, or in terms of stops, a little less than 11 (2^11 = 2048).

jrista
  • 70,728
  • 15
  • 163
  • 313
  • 3
    That 71% (1/√2) isn't quite arbitrary; it's the point where there is a 0.5 probability that an additional photon will be absorbed and change the energy state (below that value, the probability is greater than 0.5; above it is less). It's the equivalent of the "shoulder" of a film density curve, just as the SNR value limit at the other end is equivalent to "film base plus fog". Exposure values above the shoulder become increasingly non-linear, compressing the tonal scale. (Ansel Adams would have called it the difference between Zone VIII and IX.) –  Jul 27 '12 at 22:44
  • 1
    Ah, thanks Stan! As I said, I didn't have an official document to work with, and none of the resources I found could explain the 100/140 ratio. Makes total sense though, that the 71% mark is where the shoulder would begin in film. I would like to point out, however, that in the absence of a digital tone curve applied at processing time in a RAW processor...a digital sensor is a linear device. If you are computing DR with software, then I can see how the 71% might apply...however at a hardware level, I'm curious if it still does... – jrista Jul 28 '12 at 04:02
  • 1
    When using open-source RAW processing software, it is possible to load a RAW image without applying any kind of processing outside of your basic demosaicing. The nature of RAW truly becomes apparent at that point, and while a photograph tends to look eye-scratchingly dull, flat, and lifeless, there does not appear to be much in the way of a natural curve to the raw data strait off the hardware. There is probably some, but I wouldn't say anywhere close to what film naturally has. – jrista Jul 28 '12 at 04:05
  • It's linear as long as there are sufficient low-state bound electrons in the well. When they begin to become scarce (as the cell approaches electrical saturation), photons have fewer targets to hit. The curve isn't exactly a hockey stick at that point, though, it's just not as linear as it is below the shoulder; you don't get really flat until you get very near electrical saturation. –  Jul 28 '12 at 04:41
  • thank you jrista,i understand your meaning " 100/140*the maxmum saturation point equation to the saturation luminance " is this? if so, i donnot know why take 71%.could you explain it again? thank you very much. – petalse Jul 30 '12 at 03:02
  • @petalse: Well, thanks to Stan's input, it seems the ISO body managing this standard is trying to define a method of deriving dynamic range for sensors by assuming sensors have the same characteristics as film. Film, being an analog device, has a natural response curve (differs slightly with each type of film). The upper highlight range falls off with a shoulder around 71% of the films maximum saturation (which is 1/√2, to be exact). I am not sure this is really accurate for digital sensors...they are largely linear devices at a hardware level, and only take on film-like characteristics... – jrista Jul 30 '12 at 16:43
  • ...once they are processed by a RAW processing tool. RAW processors usually apply a digital "response curve" (usually called a tone curve in this respect) to give the original and rather dull looking native RAW image a more lively, saturated, contrasty look. SOME RAW processors offer some tone curves that simulate film, however these days as maximizing dynamic range is becoming more and more important, tone curves are becoming less drastic and therefor less like film's natural response curve. Given the rather arbitrary nature of RAW, I am not sure the ISO algorithm is accurate. – jrista Jul 30 '12 at 16:45
  • To be specific, without access to the actual ISO 15739 spec, I cannot be 100% certain whether "saturation luminance" refers to the sensors hardware "maximum saturation", the digital RAW files "maximum luminance level" (i.e. 2^14, or 16384), or to the 71% value of maximum saturation. A more complete resource I found over the weekend seems to indicate "saturation luminance" is the point at which a digital pixel (post ADC) reaches its maximum level, however that contrasts with other resources that seem to indicate it was the 71% value of the maximum hardware saturation. – jrista Jul 30 '12 at 16:48
  • You would really need the actual ISO 15739 spec to know for certain what it means, as there should be some explicit mathematical formulas that explain how each term is derived and how the final equation to compute DR is assembled. – jrista Jul 30 '12 at 16:49
  • “The saturation luminance is that 71% of maximum saturation.…………would be about 22000 electrons * .71, or 15620 electrons. That is a ratio of 1952.5:1 in terms of electrons, or in terms of stops, a little less than 11 (2^11 = 2048).” From this passage，the value of saturation luminance is constant to the same image sensor .if so, it is no significance to measure a cell phone camera's dynamic range through the formula.why? – petalse Aug 02 '12 at 02:19
  • I don't know what passage you are referring to... – jrista Aug 02 '12 at 06:15
  • O,it is refer to “The saturation luminance is that 71% of maximum saturation. In more detailed terms, ISO 15739 DR is the ratio between a luminance level that matches the noise ratio (say 8 electrons worth) and the 71% of the maximum saturation point, which for say Canon's 18mp sensors would be about 22000 electrons * .71, or 15620 electrons. That is a ratio of 1952.5:1 in terms of electrons, or in terms of stops, a little less than 11 (2^11 = 2048).” you mentioned above – petalse Aug 02 '12 at 07:48
2

A digital sensor has a range of values it can store - for example with 8 bits, it can store values between 0 and 255. The value 255 represents the highest (brightest) luminance value. That is what they are referring to by "receives a unique coded representation".

Anything brighter than that maximum value will also be registered at that same digital value of 255, so the luminance is "saturated" and the camera can't record any higher. The camera will report any brighter areas at the same digital value and you won't be able to distinguish them - they'll all look the same.

Community
  • 1
MikeW
  • 33,408
  • 10
  • 88
  • 123
  • Thank you Mikew, i understand you mean.But i want to know how to get this value that the "saturated luminance" when you calculate DR. In a real-life situation,for example with 8 bits, you get the value-"127" as the saturated luminance or others? how do you decide? thank you – petalse Jul 30 '12 at 02:33
  • Downvoted you by accident, can't seem to undo this until the question is edited, sorry! – Matt Grum Oct 17 '12 at 11:01