Pages: 1 2 [3]   Go Down

Author Topic: Canon Highlight Priority  (Read 81902 times)

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
Canon Highlight Priority
« Reply #40 on: December 09, 2007, 04:20:18 PM »

Quote
The confusion is in the language. I notice that John Sheehy has amended his explanations a couple of times.
[a href=\"index.php?act=findpost&pid=159417\"][{POST_SNAPBACK}][/a]

There are a lot of slippery slopes talking about this stuff.  There are a lot of issues of context, especially as far as absolutes and relatives are concerned.  The policy I try to stick to is to use relative descriptions as default, and talk about absolutes with qualification.  For example, in DPReview, I made a comment about something having less shot noise, and someone (who also happens to be on this board) replied that lower shot noise is not a good thing.  What he meant is that a lower absolute, linear amount of shot noise means that less photons are collected, but in general writing, I think we are more concerned with S:N than absolute noise, so the situation that in one context has "more absolute shot noise", has lower shot noise in the sense that we can't see it as well.

For HTP, the camera seems to meter exactly the same if you have HTP on or off; you get the same Av and Tv values in the same situation.  What is different, is that even though you have the camera set to a specific ISO setting, *internally*, the camera hardware is processing the capture as if were 1/2 the stated ISO; the gain the camera uses at ISO 200 with HTP is exactly the same gain as it uses with ISO 100 and no HTP.  If you shot with the same manual exposure for two images, one with ISO 200 and HTP, and the other with ISO 100 and no HTP, the RAW DR, noise, IQ, etc, are all the same.  They are basically the same RAW capture (as similar as any two ISO 100s or any two ISO 200s w/HTP).
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1560
    • http://www.guillermoluijk.com
Canon Highlight Priority
« Reply #41 on: December 09, 2007, 04:20:49 PM »

Quote
Someone said:
...HTP results in one stop less of exposure at the time of recording the exposure.

And Ray answered:
It doesn't. Exposure is unaltered. Only ISO setting is altered.

I think the discusion about this would come to an end if you just define what exposure is.

- If we call exposure to the optical exposure that photocaptors receive and that still has to be amplified by ISO gain, then exposure doesn't change with ISO and therefore HTP doesn't affect exposure.

- If you refer as exposure to the electronic exposure with which the ADC is fed, then exposure changes with ISO so the HTP mode would reduce exposure.

You are both right, it's simply about language. But I think you already know.
« Last Edit: December 09, 2007, 04:23:11 PM by GLuijk »
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
Canon Highlight Priority
« Reply #42 on: December 09, 2007, 04:30:51 PM »

Quote
You may see this difference as irrelevant, I don't. Of particular interest is, that ACR does not know the camera good enough, it thinks that everything over 13600 is clipped, independently of the ISO.

(Note, that the above fact makes all reviews relating the DR invalid, which use ACR for this purpose.)
[a href=\"index.php?act=findpost&pid=159525\"][{POST_SNAPBACK}][/a]

It's a real shame that converters don't have the capability to learn individual cameras with a series of test shots.  Bad pixels could be declared worthless, or their errant sensitivities taken into account.  Sensor non-linearities near saturatio could be relinearized; permanent dust/scratch shadows on the sensor could be corrected; real clipping points could be learned for different rows and columns of pixels; different amplifications of rows or columns of pixels could be noted and corrected, etc, etc.
« Last Edit: January 03, 2008, 02:50:47 PM by John Sheehy »
Logged

Slobodan Blagojevic

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 9335
  • When everybody thinks the same... nobody thinks.
    • My website
Canon Highlight Priority
« Reply #43 on: December 09, 2007, 05:11:38 PM »

Panopeeper,

You (Panopeeper) are over-complicating this very simple issue (to use your own words). This thread is about HTP and your last posts are not relevant to it, being nothing else but petulant nitpicking and hairsplitting over definitions and language usage.

As they say about pornography: hard to define, but you will know it when you see it. The same with underexposure: any photographer but absolute beginners understands perfectly well what underexposure means, regardless of how you define "exposure".

Quote
One stop lower exposure and one stop lower ISO are not the same


Regardless of the accuracy of the above statement and regardless of specific qualities of each ISO step (in terms of noise and dynamic range), one stop lower ISO is always going to result in underexposure (all other factors being equal).... and that fact (underexposure) is all it matters for the subject of HTP.

I also see you are hell-bent on teaching Thomas Knoll a thing or two about the language of photography...  good luck with it... something like lecturing the Pope on the proper language of his prayers.

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 9376
Canon Highlight Priority
« Reply #44 on: December 10, 2007, 12:34:23 AM »

Quote
(emphasis by me)

Not so. One stop lower exposure and one stop lower ISO are not the same. If you believe, that it is nitpicking, then you have not seen the related details yet..

[a href=\"index.php?act=findpost&pid=159525\"][{POST_SNAPBACK}][/a]

Well, let this be an exercise in interpretation.

Following is your emphasis on my quote:

Quote
(1) Activating HTP does not change the actual and real exposure (shutter speed). It just drops the ISO setting down a stop. In relation to the 'hidden' lower ISO setting that the camera has surreptitiously used, the shot now has the equivalent of one stop less exposure.

and below is a different emphasis by me on my own quote.

Quote
(1) Activating HTP does not change the actual and real exposure (shutter speed). It just drops the ISO setting down a stop. In relation to the 'hidden' lower ISO setting that the camera has surreptitiously used, the shot now has the equivalent of one stop less exposure.

John Sheehy has now suggested a way we might clarify this issue further with his 'shot noise' analogy which is quite interesting. The statement that high shot noise is a good thing to have, and low shot noise is something to be avoided, at first seems the wrong way round. All noise should be avoided as much as possible.

However, in an absolute sense the statement is quite true. Because shot noise is inextricably a fundamental property of the behaviour of light and varies with the square root of the number of photons arriving at the sensor, low shot noise (in terms of photon count) is indicative of a low signal level and high shot noise is indicative of a high signal level.

For a good quality image, it's always better to have a higher signal level than a lower signal level (within reason of course...no need to point out that overexposure does not necessarily result in a good image   ).

With this in mind perhaps we could distinguish between two types of exposure.

(1) Absolute exposure defined by the total number of photons arriving at the sensor.

(2) Relative exposure defined by the camera's processing of absolute exposure in relation to its ISO parameters.

Does this shed light on the matter?  
« Last Edit: December 10, 2007, 12:46:15 AM by Ray »
Logged

jani

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1620
    • yet
Canon Highlight Priority
« Reply #45 on: January 03, 2008, 01:16:01 PM »

Sorry for jumping in so late, but there's one point in the ISO peeping that seems to have been missed:

Quote
ISO   Clipping
----   --------
 100   13820
 125   16383
 160   12740

 200   16220
 250   16383
 320   12740

(...)

Note, that the 1/3 stop ISOs are not native, they are the result of a multiplication of the 1/3 lower ISO respectively a divison of the values from the 1/3 higher ISO.

You seem to have ignored the divergent clipping point for 125 ISO. This seems to indicate that there isn't simply a multiplication going on, but something else.

In every other case (250, 500, 1000 etc.), the difference between clipping point from the full stop is merely 1%, but for 100 -> 125 ISO, the difference is 18.5%.

I'm not sure whether this discrepancy is significant.

But I do hope that you've reported the ACR bug to Adobe.
Logged
Jan

Panopeeper

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1805
Canon Highlight Priority
« Reply #46 on: January 03, 2008, 06:04:09 PM »

Quote
You seem to have ignored the divergent clipping point for 125 ISO

I have not ignored it. It's easy to analyze the images with my raw analyzer,  but it is quite laborous to create a demonstration with proofs (which requires lots of co-ordinated screen shots). As the in-between ISOs are interesting only for JPEGs, I did not find it worthy to work a lot with that. To boot it, the ISO100-125 issue is particularly messy.

Quote
This seems to indicate that there isn't simply a multiplication going on, but something else.

In every other case (250, 500, 1000 etc.), the difference between clipping point from the full stop is merely 1%, but for 100 -> 125 ISO, the difference is 18.5%.

I'm not sure whether this discrepancy is significant

Sure, the ISO 100 - ISO 125 relationship is different from the others.

1. Please note, that one has to substract the black level from all values (it is a constant additive) before calculating any ratios. Its average is or close to 1024.

2. The ratio between the ranges of ISO200 and ISO160 close to 1.3. The ratio between the ranges of ISO125 and ISO100 is 1.2.

Both ratios are quite far from 1/3 stop, which is 1.26 . Accordingly, these ratios  represent no important aspect.

3. The histograms of an ISO200 and ISO160 shots are virtually equal. Note, that the histograms are relative to the actual range of pixel values, though not exactly, for example where the value range is 16220, the histogram range is 16384.

The ISO 160 values are derived from the ISO 200 values by simply a division, it appears linear. The histograms of the mapped values are virually identical (they are useful, because the black level correction and the value range are considered in those).

4. The comparable histograms of ISO100 and ISO125 shots don't behave so nicely; the ISO 125 is more stretched. Detailed analysis indicates, that the ISO 125 is not a truely linear transformation of the ISO 100 values.

5. There is another clear indication, that the +1/3 step ISOs are derived from the full-stop ISOs: there are "holes" in the ISO 125 values: red and blue do not occur with all possible values (but their holes are not together). The following seuence represents consecutive pixel values:

RGB, RGB, GB, RG, RGB, RGB, RGB, GB, RG, RGB, RGB, GB, RG, ...

There are no holes in green; I suspect, that the two different greens would show holes, but I don't care enough to go after that.

A layered TIFF can be donwloaded with the screenshots of histograms of eight shots, pairwise fromHERE

Quote
But I do hope that you've reported the ACR bug to Adobe.
Right. I posted it on the respective Adobe forum, but no-one cared to answer it.

The problem of misrepresenting the clipping point is not new, it occurs with several cameras. However, now with the 14-bit raw files and huge pixel value ranges this is more of a problem.
Logged
Gabor

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
Canon Highlight Priority
« Reply #47 on: January 03, 2008, 07:03:52 PM »

Quote
Sorry for jumping in so late, but there's one point in the ISO peeping that seems to have been missed:
You seem to have ignored the divergent clipping point for 125 ISO. This seems to indicate that there isn't simply a multiplication going on, but something else.

In every other case (250, 500, 1000 etc.), the difference between clipping point from the full stop is merely 1%, but for 100 -> 125 ISO, the difference is 18.5%.

I'm not sure whether this discrepancy is significant.

But I do hope that you've reported the ACR bug to Adobe.
[a href=\"index.php?act=findpost&pid=164826\"][{POST_SNAPBACK}][/a]

100 clips lower than you'd expect with the remaining pattern, probably because the sensor goes into saturation, or at least non-linearity, before it would reach the clipping point of 200, 400, etc.  The 20D had the same problem, but the 20D went up to 4095 in the RAW data (3967, after black point subtraction) at every ISO, and for 100, the upper highlights were somehow stretched to make them reach 4095.  You'll see figures (such as Roger Clark's) that suggest that the 20D sensor wells collect 51K photons, but the highest that gets digitized in the RAW data at ISO 100 is about 44K.

Why they just couldn't start the camera at ISO 115, or at 125 to keep things somewhat standard, is beyond me.  This obsession with 100, 200, etc does nothing but compromise cameras.  Perhaps it is marketing; people might think that not having ISO 100 is a bad thing (the fact is, an ideal digital camera with current well depth technology would have a minimum ISO of about 250).

Anyway, back to the 40D, the clipping points can vary from camera to camera, and can vary throughout the image based on pixel row and column, because, apparently, some offset correction is made based on the blackpoint by line and/or column after the data is clipped in th e highlights.  I tried to chart the relevant info about the 40D ISOs based on someone's RAW samples, and at first I tried to use the lowest clipping point in the file, but that was sometimes the value of a single line that waqs far below the mean and median, so I redid it with the mdian clipping value, instead.  Here are the details ("adjusted" refers to subtracting the blackpoint of 1024 from all of the values):

Logged

t.koetting

  • Newbie
  • *
  • Offline Offline
  • Posts: 4
Canon Highlight Priority
« Reply #48 on: January 04, 2008, 05:41:11 AM »

im trying to figure out what you are doing here

i still dont get it how you can evaluate things like well-depth without using some sort of calibrated light sorce to fill the sensor with a specific rate of photons ...

And John: if i understand right, you say it would make more sense to put the sensors low/base-sensitivity to a value where its filled to the max. ok, but fill it to the max at what light-levels and exposure-time? to me that variable needs to be defined as well, so what is the point im missing here?

thanks
« Last Edit: January 04, 2008, 05:45:01 AM by t.koetting »
Logged

jani

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1620
    • yet
Canon Highlight Priority
« Reply #49 on: January 04, 2008, 07:58:51 AM »

Quote
I have not ignored it.
Okay, but you didn't mention the divergence in your post, so it did seem that way.

Thanks for the clarification, though; it helps!
Logged
Jan

jani

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1620
    • yet
Canon Highlight Priority
« Reply #50 on: January 04, 2008, 08:16:27 AM »

Quote
100 clips lower than you'd expect with the remaining pattern, probably because the sensor goes into saturation, or at least non-linearity, before it would reach the clipping point of 200, 400, etc.  The 20D had the same problem, but the 20D went up to 4095 in the RAW data (3967, after black point subtraction) at every ISO, and for 100, the upper highlights were somehow stretched to make them reach 4095.  You'll see figures (such as Roger Clark's) that suggest that the 20D sensor wells collect 51K photons, but the highest that gets digitized in the RAW data at ISO 100 is about 44K.

Why they just couldn't start the camera at ISO 115, or at 125 to keep things somewhat standard, is beyond me.  This obsession with 100, 200, etc does nothing but compromise cameras.  Perhaps it is marketing; people might think that not having ISO 100 is a bad thing (the fact is, an ideal digital camera with current well depth technology would have a minimum ISO of about 250).
This is quite interesting, since the 20D's "100 ISO" setting is really equivalent to 125, while the 40D's "100 ISO" is at least very close to 100.

It would appear then, that the 20D ought to have started at 160 and the 40D at 125.

What you're pointing at may explain why Nikon usually only offers standard sensitivities from 200 ISO and upward, with 100 ISO as an extension.

Quote
Anyway, back to the 40D, the clipping points can vary from camera to camera, and can vary throughout the image based on pixel row and column, because, apparently, some offset correction is made based on the blackpoint by line and/or column after the data is clipped in th e highlights.  I tried to chart the relevant info about the 40D ISOs based on someone's RAW samples, and at first I tried to use the lowest clipping point in the file, but that was sometimes the value of a single line that waqs far below the mean and median, so I redid it with the mdian clipping value, instead.  Here are the details ("adjusted" refers to subtracting the blackpoint of 1024 from all of the values):
Unless I'm misreading the numbers, using 160 instead of 200, 320 instead of 400, 640 instead of 800 and 1250 instead of 1600 may be quite acceptable in terms of S/N and DR. Fascinating.

It would be interesting to see how this compares to the 1D(s) MkII, MkIII, 5D and 30D.
Logged
Jan

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
Canon Highlight Priority
« Reply #51 on: January 04, 2008, 08:49:40 AM »

Quote
This is quite interesting, since the 20D's "100 ISO" setting is really equivalent to 125, while the 40D's "100 ISO" is at least very close to 100.

It would appear then, that the 20D ought to have started at 160 and the 40D at 125.

Well, the quantum efficiencies of the two cameras are very close, much closer than that, so there must be some other factor.

Quote
What you're pointing at may explain why Nikon usually only offers standard sensitivities from 200 ISO and upward, with 100 ISO as an extension.

It might explain the D3, but there are also Nikons with relatively low quantum efficiency that start at 200, so those might have a small fill factor but a large microlense.

Quote
Unless I'm misreading the numbers, using 160 instead of 200, 320 instead of 400, 640 instead of 800 and 1250 instead of 1600 may be quite acceptable in terms of S/N and DR. Fascinating.

It would be interesting to see how this compares to the 1D(s) MkII, MkIII, 5D and 30D.
[a href=\"index.php?act=findpost&pid=164974\"][{POST_SNAPBACK}][/a]

The 1D cameras and the 5D use an analog amplifier, so all their ISOs go into the ADC proportional to the ISO, and they don't have the clipping variations of the 30D and 40D (but they are always amplified from the lower ISO, so the 160/320 group have the peak read noises).

The 30D is similar to the 40D in approach, but the 160 group have a DR advantage over the 100 group a little bit greater than with the 40D.
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
Canon Highlight Priority
« Reply #52 on: January 04, 2008, 08:58:41 AM »

Quote
im trying to figure out what you are doing here

i still dont get it how you can evaluate things like well-depth without using some sort of calibrated light sorce to fill the sensor with a specific rate of photons ...

The number of photons at full-well is not directly related to exposure.  Depending on the camera, only a certian percentage of photons can be captured, and others reflected or turned to heat.

Quote
And John: if i understand right, you say it would make more sense to put the sensors low/base-sensitivity to a value where its filled to the max. ok, but fill it to the max at what light-levels and exposure-time? to me that variable needs to be defined as well, so what is the point im missing here?
[a href=\"index.php?act=findpost&pid=164964\"][{POST_SNAPBACK}][/a]

That's a characteristic of the sensor.  Of course it needs to be calculated.  The point is, the camera should have a low ISO that combines maximum photon capture with minimum read noise relative to max photons.  Many cameras fail to bring these two qualities together to do this.
« Last Edit: January 04, 2008, 08:59:32 AM by John Sheehy »
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1560
    • http://www.guillermoluijk.com
Canon Highlight Priority
« Reply #53 on: January 13, 2008, 02:51:41 PM »

Although I think that reading the article is clear that HTP is nothing more than shifting ISO down by 1 f-stop and shooting at the same shutter/aperture, I have a real sample: 2 shots with camera settings:
- Up: ISO200+HTP, f8 1/40s
- Down: ISO100 no HTP, f8 1/40s

Both images have been developed with neutral and identical parameters using DCRAW, and the final curve applied was also the same for both images.

This was the scene: same appearance




These are demosaiced but non-WB histograms: both display same exposure




These are 100% crops: twins (except obviously for the noise distribution)




The usefulness of HTP is only in JPEG, where the camera will ensure a correct exposure in the mid tones (not underexposed because of the ISO reduction) but prserving the highlights. Nothing that could not be achieved shooting RAW with 1 f-stop ISO down and same shutter/aperture params and a proper post processing.
« Last Edit: January 13, 2008, 02:55:12 PM by GLuijk »
Logged
Pages: 1 2 [3]   Go Up