I would not trust a Tony Northrup video for technical discussion, he tends to give controversial facts only to get more traffic.
Yes, in the meantime I found out, that definitely not all cameras are " iso invariant" (see above). In that respect his video is certainly nonsense.
1 Like
It’s been a while since I’ve seen the video. But if I remember correctly the main point Tony tries to get across is that for example ISO 100 on one camera is (potentially) completely different from ISO 100 on another camera. Since there doesn’t seem to be a standard on ISO anymore he concludes that “ISO is totally fake”.
Yes, that is indeed one of his points. But he has also other points. At 7:00 minutes on he makes a comparison for a nikon camera (iso 400 vs. iso 25600) to show that it is “isoless”. Right afterwards he claims that this is the case for most cameras today, which I think, is wrong.
This link from Thomas_Do in german language is very interesting with many deep insights.
The author’s conclusion:
I claim that in 85% of the cases of photography, images of identical quality can be produced with conventional camera sensors and correct camera settings. For 5% you can also underexpose 1-2 apertures with old sensors and then brighten them up again in the software, as well as optimize the image with today’s very good noise reduction (if necessary even extreme problem cases with masks succeed). For another at least 5% there are already series exposures and HDR(I) software available today.
Whether the acquisition of a new camera with such a sensor is worthwhile for the remaining 5% of cases occurring in practical photography depends on your activity:
I advise hobby photographers against an immediate new purchase only for this reason.
However, professional photographers who have to make a living and are paid to deliver high-quality photos for DIN A3 magazines and demanding customers at all times, even under critical and extreme lighting conditions, are highly recommended. For this group of people, even high four-digit editions are not difficult.
This fits to my personal situation. I do around 95% of my shots at ISO 100 which gives me only a ‘tiny’ noise that I can remove easily within Darktable (APS-C sensor from 2013 on my Pentax K-50).
With a newer bigger sensor I could do maybe around 98% of my shots at ISO 100 with a ‘even more tiny’ noise.
But I doubt I would see a big difference after denoising in the RAW processor.
It depends on how pedantic you want to be. Certainly ISO has become less consistent among cameras with the advent of digital cameras. Are we comparing untouched raw data or (profiled) denoised images?
Hello,
let me try to clarify the following: There is no way to “amplify” the signal from a CCD (or CMOS) detector. The quality of the signal (i.e. the signal-to-noise-ratio SNR) is determined solely by the number of photons detected and the electronic noise of the system. Neglecting the latter for a good camera and average usage, the number of photons collected per pixel is only determined by the exposure time, the aperture and the pixel size.
I am not familiar with the technical details of commercial cameras but I know how astronomical CCD-cameras work and there should be no fundamental difference. The number of photons collected is equivalent to a voltage, measured after the end of the exposure for each pixel leading to the signal. This voltage is converted to a digital number by the ADC (analog-to-digital-converter). What then happens, depends on the application. And here enters according to my understanding the ISO number, which I feel is sort of a misnomer for historical reasons. In astronomy we call it the gain. Depending on the digital representation of the electronics, let me assume a 2 bytes per channel and pixel, which corresponds to digital numbers between 0 and 65535. The ISO setting (or gain) determines, how the voltage is mapped into this available number range. For high exposer levels (high brightness, long exposure time) the scaling factor from voltage to digital number needs to be low, in order to map a large number of photons to the maximum available number of 65535. For low levels, the conversion factor is high, stretching a smaller range of photons to the same numerical range 0 to 65535. But the latter also means that with this scaling factor you also scale-up the noise!
According to my understanding, a high ISO just better samples the numerical range, so it is only a digital effect. It does not change the measured signal, i.e. the number of photons detected, which in turn determines the SNR.
I do not see, what this has to do with JPG or TIF, as stated above.
Hermann-Josef
3 Likes
I may have investigated this before but need a reminder:
If we clip any pixels (or don’t), which is better?
- Gain via ISO, or
- gain via post-processing?
If we know the ISO gain and apply it in post instead, would we get the same result?
@afre
Without having tried explicitly I would say that there should be – except for numerical differences – no difference in terms of SNR. With higher ISO you just spread the signal over a larger range in numbers. This is, how I see it.
Hermann-Josef
It is possible that the gain is nonlinear.
I think you are right, as far as I researched. This would explain while you see less noise in long-time exposures.
Wikipedia goes into that direction, too.
After reading this I don’t think that ISO has something to do with the ‘sensitivity’ of the sensor (like it was on analog films).
Considering this ISO would be just a multiplicator of the sensor data which means it is an effect.
Cannot tell if it’s linear or not. But it helps me getting the exposure times short when there is movement in my picture or there is not much available light.
I did some digging for sensor data sheets and such; one I found does apply gain before the ADC, but from a binary number:
See page 16.
In that case, the gain resolution is 32-bit, which still introduces quantization, but at a rather fine-grained level.
Thinking out loud, but the base ISO of a camera probably is about the sensor’s sensitivity.
You’re right, should have written: changing the ISO on a camera does not change the sensor’s sensitivity 
1 Like
Try it and see. For my camera (Nikon D800 DSLR, full-frame, getting a bit old now) there is a difference. “Gain via post-processing” gives more colorful noise, ie there is a wider variation in chroma, than by varying ISO. To my eyes, the post-proc gain looks far noisier.
I also suspect (though haven’t formally tested this) that gain via post-processing loses more shadow detail (ie more banding) than gain via ISO.
Just saw a thread by one Pentax K-1 owner who tried the same with a different result:
But I think that the demosaicing method does have its impact, on my camera I see more noise with PPG than with AMaZE.
Maybe the ‘better’ noise performance depends on the camera (sensor) model or the demosaicing method, maybe on the user’s taste
@snibgo
Here is the test with my SonyF828, albeit a rather old camera. On the left ISO 64 on the right ISO 800, aperture and exposure time being kept the same:
I see practically no difference.
As I said, there is no such thing as a physical amplification in CCDs (in contrast to image tubes). Thus all one can do is to map the signal range to the numerical range, which is what the ISO setting does according to my understanding. The noise is set by the number of photons detected and nothing else (except readout noise, which I assume to be small). Here ISO cannot make any difference.
Hermann-Josef
And I doubt there is a practical one. The right ISO 800 image seems to be a bit more saturated.
Assuming you increased the exposure of the left ISO 64 image in a program, how does its algorithm work? If it only increases a luma channel then it would explain the difference.
Would it be the same result if you use another program or another method (brightness, exposure, gamma or whatsoever)?
If you’re precise, both images are not comparable.
But in reality you would just move a ‘saturation’ slider some more pixels in the left picture and won’t have a remarkable difference.
I just set black and white point in the histogram, which was perhaps not completely identical for both images. Thus the appearance of a slight difference in saturation. That was all for both images. So I did not change the pixel values – in contrast to the procedures you mention (brightness, exposure, gamma …)!
Concerning the issue with ISO I want to quote from the book by Tooms (2016): “Colour reproduction in electronic imaging systems”: On page 407 he writes: “The image sensor will have an opto-electro-conversion factor (OECF), which is normally linear and will output its signal to variable gain amplifiers which are adjusted for minimum gain, consistent with providing a standard level signal after the desired adjustments to the iris and shutter speed exposure settings. The gain adjustment is usually calibrated in terms of the ISO rating, which in turn is related to the ASA film sensitivity or speed rating familiar to traditional photographers.”
This confirms that the signal is just multiplied by a gain factor, which in turn does not change the SNR.
I also found a webpage, confirming what I just wrote. Scroll down to the part about “Common ISO Myths and Misconceptions” and read the paragraph “Is ISO “Sensor Sensitivity”?” The fundamental difference to the ISO/ASA of a photographic film is, that a film with high ISO does indeed have a higher sensitivity. If you raise the ISO setting of your camera, it does not increase the sensitivity of the detector.
However, one never knows what the firmware in the camera does in detail, if one raises the ISO setting. Thus it may well be, that due to some clever algorithm the image with high ISO setting does look somewhat better than what you get with a lower ISO setting and some image processing with the software of your choice.
Hermann-Josef
I suspect the SNR comparison depends on the camera model (which includes the physical sensor, any analogue gain, the ADC, any digital gain, and whatever post-processing is done. Especially any noise-reduction, obviously.
I don’t know exactly how my camera works. When I twiddle the ISO dial, what changes? Is there a change in analogue gain, or what?
This is somewhat frustrating, but I never fully understood how film worked, even just B&W, let alone colour. I think my D800 uses analogue gain for different ISO, up to the “extra” ISO called H1, H2 etc which are digital. But I could be wrong.
Personally, I’m not much bothered about what the theory says should happen. Instead, I look at what actually does happen. For example, an out-of-focus gray card, under office lighting (LED), constant exposure 1/800s f/2.8, at ISO 6400 and ISO 100:
set AREA=3500 2500 300 300
set CROP=-crop 300x300+3500+2500
%DCRAW% -v -A %AREA% -6 -o 1 -T -O 4398.tiff AGA_4398.NEF
%DCRAW% -v -A %AREA% -6 -o 1 -T -O 4402.tiff AGA_4402.NEF
%IMG7%magick 4398.tiff -strip %CROP% -auto-gamma +write isonse_4398.png -colorspace HCL -format "%%[fx:mean.g]\n%%[fx:standard_deviation.g]\n" info:
%IMG7%magick 4402.tiff -strip %CROP% -auto-gamma +write isonse_4402.png -colorspace HCL -format "%%[fx:mean.g]\n%%[fx:standard_deviation.g]\n" info:
The numbers are:
ISO 6400:
0.156747
0.0983761
ISO 100 (given post-proc gain):
0.170328
0.132537
The numbers show that post-proc gain has higher saturation (mean chroma), and a wider variation (standard deviation of chroma).
In real photography, I use dcraw’s noise reduction for high ISO, which improves those images.
So, if post-proc gain is noisier, do I avoid it? No. On the contrary, I generally set the camera to auto-expose with exposure compensation up to -2 stops. I deliberately underexpose to avoid clipping highlights, then correct with post-proc gain. For me, some extra noise is a price worth paying to avoid clipping.
2 Likes
This is what is happening, according to what I understand and read. However, this does not change the SNR, which is the main issue.
What do you mean by that?
I think one should be careful in the discussion to separate theory and praxis. My emphasize was on theory in order to understand what is going on. The practical part are the personal preferences.
If the variance in the higher ISO image is lower, then this could well mean, that the firmware is applying some algorithm to reduce noise (e.g. wavelet filtering). I doubt that the manufacturers will tell us exactly what they do.
Hermann-Josef