Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: Out of Focus Highlights not smooth?  (Read 7299 times)
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1290



WWW
« Reply #20 on: December 09, 2009, 04:35:27 PM »
ReplyReply

Quote from: PierreVandevenne
At the root, we have a photon counter: it doesn't really matter much if we multiply/divide at the AD converter, in the camera, or in software post-processing
It does matter if you multiply in the analogue domain (electronic amplification) vs in the RAW domain (once data has been converted to digital).

Analogue amplification (genuine ISO) improves signal to noise ratio; software amplification (fake ISO) doesn't.

2 shots taken at T=1/4s, f/13, 50mm (note that the same aperture and shutter speed mean the same amount of photons entering the camera), just ISO was changed between them and exposure adjusted in pp to match on both images:



The higher the genuine ISO for a given amount of light hitting the sensor, the higher the SNR.
« Last Edit: December 09, 2009, 04:40:21 PM by GLuijk » Logged

PierreVandevenne
Sr. Member
****
Offline Offline

Posts: 510


WWW
« Reply #21 on: December 09, 2009, 05:12:47 PM »
ReplyReply

Quote from: GLuijk
It does matter if you multiply in the analogue domain (electronic amplification) vs in the RAW domain (once data has been converted to digital).

Analogue amplification (genuine ISO) improves signal to noise ratio; software amplification (fake ISO) doesn't.

2 shots taken at T=1/4s, f/13, 50mm (note that the same aperture and shutter speed mean the same amount of photons entering the camera), just ISO was changed between them and exposure adjusted in pp to match on both images:

The higher the genuine ISO for a given amount of light hitting the sensor, the higher the SNR.

I hear you but I was talking about electron counts as real raw data. Obviously, at 100 ISO, in this case, your electron count is too low so you aren't in the optimal zone I was talking about.

Signal to noise ratio is very simply the ratio of the signal you are measuring vs all the possible sources of noise
http://learn.hamamatsu.com/articles/ccdsnr.html

All other things being equal, and with sufficient well capacity, if you defined "genuine iso" as the QE of the sensor, yes, SNR increases.

EDIT: if you aren't convinced by what I am saying, check the DXO labs analysis for different cameras and you will clearly see that there is one "sweet" spot for dynamic range and SNR (usually in the low "ISO" range) and that things go downhill for virtually all cameras from there as "ISO" increases. Analogue amplification doesn't improve things, far from it.

http://www.dxomark.com/index.php/eng/Image...abase/Nikon/D3X
around 80 ISO

http://www.dxomark.com/index.php/eng/Image.../EOS-5D-Mark-II
around 75 ISO
« Last Edit: December 09, 2009, 05:23:58 PM by PierreVandevenne » Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #22 on: December 09, 2009, 06:04:12 PM »
ReplyReply

Quote from: PierreVandevenne
Obviously, at 100 ISO, in this case, your electron count is too low so you aren't in the optimal zone I was talking about
Low counts. i.e. deep shadows are normal parts of an image.

Quote
All other things being equal, and with sufficient well capacity, if you defined "genuine iso" as the QE of the sensor, yes, SNR increases
Guillermo's genuine ISO is one, which is achieved by analog gain, not by numerical manipulation in the digital realm - this is clearly expressed in hos post.

Quote
check the DXO labs analysis for different cameras and you will clearly see that there is one "sweet" spot for dynamic range and SNR (usually in the low "ISO" range) and that things go downhill for virtually all cameras from there as "ISO" increases. Analogue amplification doesn't improve things, far from it
This is a misinterpretation of the analysis. Analog amplification does improve "things", it clearly increases the SNR (up to some camera dependent limit, for the high analog ISO steps not always improve the SNR).

The practical problem is, that increasing the ISO by 1 EV roughly doubles the pixel values. Going from ISO 100 to 1600 would require four bits more; that is probably too much for the A/D electronics (caution, I am incompetent on the HW domain). Thus, if the bit depth is for example 14, increasing the ISO from 100 to 200 would create 15 bit values; the top of that gets thrown away. The higher ISO enhanced the SNR, but it eliminated one stop from the highlights; therefor the dynamic range is reduced by the difference between one stop and the enhancement.

Look at the noise/DR graphs of the Canon 5D2; the difference between ISO 100 and 200 is very small, but from 1600 to 3200 the loss is practically a full stop (some copies of the 5D2 are somewhat better, they show a slight enhancement from 1600 to 3200 as well).

Canon 5D2 Noise/DR
Logged

Gabor
Derryck
Full Member
***
Offline Offline

Posts: 170



WWW
« Reply #23 on: December 09, 2009, 07:31:57 PM »
ReplyReply

Keith from Northlight Images did an iso test with the 7D and it shows how using the in-between iso settings can result in higher noise than the full stop iso settings. 7D iso test

Derryck.
Logged

Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #24 on: December 09, 2009, 08:14:38 PM »
ReplyReply

Quote from: Derryck
Keith from Northlight Images did an iso test with the 7D and it shows how using the in-between iso settings can result in higher noise than the full stop iso settings
Keith posted this link on another thread; I looked at the method and found it meaningless.

There is an important point to understand when discussing this subject: the noise with different ISOs must be compared on patches with equal light capture. Most photogs don't understand this.; they compare two shots with different ISO and with vastly different light capture. When increasing the ISO by one stop, the metered exposure decreases by one stop. The ISO increase increases the SNR, but the reduced light capture decreases it more; thus the higher ISO seems to yield higher noise, and the users attribute that to the higher ISO.

The meaning of this regarding the 1/3 ISO stops is, that even though the ISO gain is the same for example for ISO 200 and 250; or for ISO 320 and 400, the exposure is different. For example make a shot with ISO 400, metered exposure. Then make a shot with ISO 320, metered exposure. The ISO used in the 320 shot is ISO 400, but the exposure is higher, thus the noise is less - most users interpret this as ISO 320 is "better than ISO 400". On the other hand, ISO 250 is in fact ISO 200, but the exposure is 1/3 EV lower, i.e. it shows higher noise - again, due to the exposure, not to the ISO.

The study The Source of Noise contains detailed exlanations and demonstrations of these issues.
Logged

Gabor
fike
Sr. Member
****
Offline Offline

Posts: 1373


Hiker Photographer


WWW
« Reply #25 on: December 10, 2009, 10:05:12 AM »
ReplyReply

This is a fascinating conversation.  I have learned lots.

This issue has made me more careful about the old digital mantra ETTR (expose to the right), particularly when people say it is okay to see a bit of clipping in highlights because you should be able to recover it with RAW.  What you can easily overlook is how much your RAW processing can recover and, more importantly, whether  one particular color channel is clipping substantially more than others.  Clipping in the blue channel, for example, can ruin your ability to recreate a good smooth sky, though the rest of the image looks fine.  

I was out in the snow last weekend and these issues were on my mind.  I have started to actually pull back from aggressive ETTR because the other consequence is that you end up blowing out skies and clouds in the effort to make the snow white.  I'll take a good bell curve histogram any day, and I will try to shift it to the right, but not so much that the line approaching zero is off the edge.  Further, you can easily miss the fact that there is a spike just outside the right border of the JPG preview histogram.
Logged

Fike, Trailpixie, or Marc Shaffer
marcshaffer.net
TrailPixie.net

I carry an M43 ILC, a couple of good lenses, and a tripod.
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #26 on: December 10, 2009, 11:13:20 AM »
ReplyReply

Quote from: fike
This issue has made me more careful about the old digital mantra ETTR (expose to the right), particularly when people say it is okay to see a bit of clipping in highlights because you should be able to recover it with RAW.  What you can easily overlook is how much your RAW processing can recover and, more importantly, whether  one particular color channel is clipping substantially more than others
The problem with looking at the clipping indication on the camera is, that you don't know if the clipping occured on raw level, leading to incorrect pixel values, or it is the creature of the raw conversion, caused by WB application, color space conversion and adjustments. Only a neutral setup in the camera helps by contributing to more or less raw-like histograms and clipping indication.

The so-called recovery function of ACR, controlled by the slider, does not recover anything, which was really lost. ACR - and other raw converters - often indicate clipping, when the raw data was in fact very far from clipping, but the resulting color falls outside of sRGB. Reducing the intensity by "exposure" or "recovery" brings the color into the sRGB gamut.
Logged

Gabor
fike
Sr. Member
****
Offline Offline

Posts: 1373


Hiker Photographer


WWW
« Reply #27 on: December 10, 2009, 11:30:54 AM »
ReplyReply

Yes, by recover blown highlights, I wasn't really referring to the recovery slider.  I generally don't find it helpful.  Exposure coupled with Brightness seems to have better results.  Better than those is to do two different raw outputs and then blend them with auto align layers and masking techniques. I even like this better than, so called, pseudo HDR using single RAW frames processed with different exposures.

One technique I have been enjoying some success with is to decrease the luminance and increase the saturation of only the blue channel in ACR.  Generally for landscape, you don't see this create unpleasant blue casts elsewhere in the image because there aren't many natural things that are made with the color blue, except for perhaps sky, water and ice.  In those cases, you probably are okay with tweaking the blue levels.  The only possible problem with this, and it is related to the original poster, is when you can create some posterization by pushing the blue channel too far out of synch with the other color channels.
Logged

Fike, Trailpixie, or Marc Shaffer
marcshaffer.net
TrailPixie.net

I carry an M43 ILC, a couple of good lenses, and a tripod.
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1290



WWW
« Reply #28 on: December 10, 2009, 01:34:08 PM »
ReplyReply

Quote from: PierreVandevenne
Analogue amplification doesn't improve things, far from it.

So lucky me for having a Canon 350D with such magical features.

Regards.
Logged

keith_cooper
Sr. Member
****
Offline Offline

Posts: 448



WWW
« Reply #29 on: December 10, 2009, 02:54:33 PM »
ReplyReply

Quote from: Panopeeper
Keith posted this link on another thread; I looked at the method and found it meaningless.
Thanks mate :-) - meaningless for you in respect of your different criteria I'd say...

I still stand by the results in the context they were given - they show variations in noise, as would be found in real world files processed the way most people process them.

I make no claims as to how various hardware and software does things between light going into the lens and ACR producing an image for me. Sometimes this is relevant, other times the process might as well be carried out by smart imps inside the camera and my computer.

Just depends why you are looking at things.
Logged

Wayne Fox
Sr. Member
****
Offline Offline

Posts: 2847



WWW
« Reply #30 on: December 10, 2009, 08:00:07 PM »
ReplyReply

Quote from: GLuijk
If you want to know what information is reliable, produce your own information. There is no mistery in extracting the RAW data and analysing it.

I'm not completely clear how to do this ... take two identical images at the two ISO's and then examine the RAW histogram?  I assume doing this in most raw processors wouldn't work since they are applying a correction based on the ISO (thus faking it).  Is this the right steps and which program would you use?

Is ISO 50 on the PhaseOne p65+ real?  I've assumed it was since they recommend using it when you can.
Logged

Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #31 on: December 10, 2009, 09:39:54 PM »
ReplyReply

Quote from: Wayne Fox
I'm not completely clear how to do this ... take two identical images at the two ISO's and then examine the RAW histogram?
Exactly - but you have to look at the raw channel histograms.

Example: the 5D2, a set of shots with different ISOs from Imaging Recources; controlled illumination, exposure steps corresponding to the ISO steps. The histograms from ISO 400, 200 and 100 are practically identical, for doubling the exposure and halving the ISO are "compensating" for each other. But watch the ISO 50 shot: the histogram shows the 1 EV higher exposure, because the ISO gain is the same as with ISO 100.

ISO 400
ISO 200
ISO 100
ISO   50

Quote
Is ISO 50 on the PhaseOne p65+ real?  I've assumed it was since they recommend using it when you can
I don't have raw files from the P65+, but I am pretty sure ISO 50 to ISO 800 are real, like with the P45+. However, "real" is not always "effective"; the highest real ISO gains of several cameras do not reduce the noise the least, they reduce the dynamic range by one full stop.

On the other hand, several other MFDBs, for example the P25+ don't have higher ISO than 100 (I guess they have real ISO 50 as well). ISO 200 and above are identical to ISO 100 (unlike the faked ISO gain of Canons and Nikons), "making true" the ISO is the task of the raw processing.
Logged

Gabor
Wayne Fox
Sr. Member
****
Offline Offline

Posts: 2847



WWW
« Reply #32 on: December 11, 2009, 02:10:00 AM »
ReplyReply

Quote from: Panopeeper
Exactly - but you have to look at the raw channel histograms.

Example: the 5D2, a set of shots with different ISOs from Imaging Recources; controlled illumination, exposure steps corresponding to the ISO steps. The histograms from ISO 400, 200 and 100 are practically identical, for doubling the exposure and halving the ISO are "compensating" for each other. But watch the ISO 50 shot: the histogram shows the 1 EV higher exposure, because the ISO gain is the same as with ISO 100.

ISO 400
ISO 200
ISO 100
ISO   50


I don't have raw files from the P65+, but I am pretty sure ISO 50 to ISO 800 are real, like with the P45+. However, "real" is not always "effective"; the highest real ISO gains of several cameras do not reduce the noise the least, they reduce the dynamic range by one full stop.

On the other hand, several other MFDBs, for example the P25+ don't have higher ISO than 100 (I guess they have real ISO 50 as well). ISO 200 and above are identical to ISO 100 (unlike the faked ISO gain of Canons and Nikons), "making true" the ISO is the task of the raw processing.

Thanks. Unfortunately your links aren't working.  this is quickly getting a little over my pay grade

So which program is the best to allow me to see raw channel histograms?
Logged

francois
Sr. Member
****
Offline Offline

Posts: 6819


« Reply #33 on: December 11, 2009, 04:13:17 AM »
ReplyReply

Quote from: Wayne Fox
…So which program is the best to allow me to see raw channel histograms?
Rawnalyze should do the trick ( http://www.cryptobola.com/PhotoBola/Rawnalyze.htm ).
Logged

Francois
Wayne Fox
Sr. Member
****
Offline Offline

Posts: 2847



WWW
« Reply #34 on: December 11, 2009, 12:42:37 PM »
ReplyReply

Quote from: francois
Rawnalyze should do the trick ( http://www.cryptobola.com/PhotoBola/Rawnalyze.htm ).

Doesn't appear to support files from PhaseOne backs.
Logged

Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #35 on: December 11, 2009, 01:02:24 PM »
ReplyReply

Quote from: Wayne Fox
Doesn't appear to support files from PhaseOne backs.
Convert the raw file in mosaic DNG format (for example with the Adobe DNG converter), then it should work. Linear DNGs, which are demosaiced, are not supported by Rawnalyze (those are not raw any more).

Quote
Unfortunately your links aren't working
The site is blocked as the moment. Someone or something reported some "malicious content" without specifying, which file, and the service provider blocked everything. I am in the "process of resolution".
Logged

Gabor
Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad