A few technical and theoretical data on ISO (sensitivity)
Light through a lens arrives on a camera sensor. On a sensor there are microlenses which focus light from the bigger square at photo cells. Photo cells convert photons in an electrical charge which collects depending on duration and intensity of lighting. The “superfluous” are led out from the drive on special tracks (drainage) in order not to create additional distortions of light sampling of light.
On the picture are schematically provided microlenses and photo cells, under them. in this case there is reduction of barriers between microlenses in new cameras that should increase sensitivity of a matrix.
There are 3 elements which are reading out each color (R, G, B) on the conditional pixel (under one lens). In some matrixesthere are more of them.
Charge of a separate element amplify its personal preamplifier (CMOS matrix).
The camera makes every time 2 pictures. One with an open lock to capture the image, and another with closed to shoot a picture of parasitic charges on preamplifiers which the camera will subtract from the first picture and thus will strongly reduce influence of non-uniformity of preamplifiers characteristics (as much, as pixels).
Otherwise was on CCD matrixes, where there was one preamplifier on all and the diagram was very inflexible.
Further data on a photo cells charge will be transformed to numeric data by means of camera ATsP with digit capacity 12 or at most 14 bit.
Digit capacity of ATsP provides us measuring accuracy of an analog signal and its conversion to numerical value. And there is a certain threshold value of tension for measurements, concerning which the charge is measured. If it is small, it allows to measure a charge very precisely in case of high ISO, but on the other hand here already all diagram shall provide such measurements, differently you risk to receive noise level higher than a level of measurements (bad SNR).
If the voltage is high, we can’t work at high ISO (in case of a low signal) i.e we won’t be able to measure it. The accuracy will be insufficient.
If tension is average, we can envelop sufficient dynamic range, and to calculate mean values for which there is no digit capacity of ATsP.
These values ( ISO) are already measured and known for some cameras.
“Dishonest” ISO values can turn out both a signal preamplifier, and simple digital interpolation.
The ratio of preamplifier influence and digital processing isn’t known to me and has rather scientific value, unlike known “honest” ISO.
As far as I note, Nikon (or it is more correct to speak about the vendor of Sony sensors) already began to use the adaptive regulation of preamplifiers, as it in the case of ISO increase the truncating of light is noticeably decreased and favourably looks when comparing pictures from the cameras Canon and Nikon. The response from Canon, I think, is not far.
What important parameters influence on perception of noise in the photo?
1. Number of noise in comparison with the useful pure signal (SNR — signal to noise ratio).
2. Nature of noise. Noise happens in the different form and color. The figure of noise and its color are perceived by our eyes / brain in a different way.
On what these two previous points depend?
1.The number of noise depends from:
— sufficiency/insufficiency of scene lighting. It is less than light — more noise in DARK areas. I recommend to pay attention that in light areas can be very little or generally not to be noise on high ISO and, on the contrary, on ISO 100 in dark areas there is a lot of noise.
— Bad ATsP. Incorrect processing of a signal from photo cells, parasitic charges etc. The vendor fight against it, but it isn’t too active as this type of improvings can strongly influence on the camera price (compare the price of the home amplifier and high-quality, to very low SNR).
— Low sensitivity of a matrix (in turn depends on sensitivity of photo cells, their layout, microlenses and barriers between photo cells). I find it difficult to tell that big improvings are hindered on this point, but by the fact that sensitivity of matrixes increases extremely slowly.
— Heating of a matrix by the long operation. With transition to CMOS technology of matrixes the question became less sharp, but is all the same actual for astronomers and shootings on great optical exposure.
2. Nature of noise depends from:
— ATsP devices
— Structures of a matrix and layout of photo cells on it
— The filter located before a matrix
Camera Canon 1D X
Generally the principle of 1D cameras is familiar to me on the camera Canon 1D mark II N which honestly served me for many years. The principle consists in uncompromising quality for the prof. cameras which are only 1D cameras at Canon. For this reason and it is interesting to compare noise of the 1D X to noise on the camera Canon 5D Mark II which belongs to an amateur segment of cameras. It is not bad that Canon 5D is the amateur camera. On the contrary, it is very good as amateur camera and therefore many users use it professionally, for commercial shooting. Including me.
Earlier I already published the article Digital noise (comparing of Canon 1d Mark II N, Canon 20D, Canon 5D, Canon 5D mark II where is visible that there was any big advantage in the previous generations, contrary to announcements of marketing specialists.
We will look, whether there are improvings in the modern top camera.
The camera on a tripod, the lens Carl Zeiss Makro-Planar 50/2 ZE, an F8 diaphragm.
At first I tried to add to the correct exposure of Canon 1D X 2 steps, to lighten dark areas and to make visible noise, and then to deliver the same settings in Adobe Camera RAW for the Canon 5D Mark II. But look at result.
The white balance is identical, a zero drive is disconnected in the camera and in the converter.
According to pictures it is obviously visible that ISO100 is understood differently by cameras. The picture in Canon 1D is more darkly because of an overexposed photo, with the Canon 5D Mark II it looks absolutely badly, with big overexposure.
In general it also doesn’t surprise me. I felt it long ago. And not I one. The same DxO publishes a ratio declared by the vendor of the ISO camera and really measured.
It was necessary for the correct test to set up the identical brightness of a photo on digits for each photo that in identical places they matched on brightness as much as possible.
It isn’t necessary to take literally i.e. these colors are conditional one, and I needed to lead pictures to identical brightness. Black and white depended completely on settings in the converter.
These pictures are uder-exposure and later I made them in grey tone to receive more noise which we will analyze.
I noted earlier that different vendors understand ISO in a different way but that at one vendor they differed, I saw it for the first time.
It appeared, that the picture with Canon 1D X is more dark than a picture with the Canon 5D Mark II approximately on 0.55 EV. Plus in case of identical balance white pictures differ on color gamma. The picture with the Canon 5D Mark II has more red shades, and the picture with Canon 1D X is more reliable (I remind that the identical white balance is 4450K, +49).
For alignment of brightness the pictures correction were +0.55 EV for a picture with Canon 1D X.
In general pictures are clarified as:
Canon 5D mark II: +2 EV
Canon 1D X: +2.55 EV
Data of noise without brightness correction in the converter:
Canon 5D mark II, SNR=21.78/1.45=15.02
Canon 1D X, SNR=20.76/1.07=19.40
Measurements are taken on a big uniform section of the sky. In this case on 1D X there is much less noise.
Canon 5D mark II
Here, for example, 100% crop of pictures from the Canon 5D Mark II on ISO 3200 and ISO 4000 with one exposure 1/2s, a diaphragm as well as above is F8.
For alignment of brightness I lifted value of exposure in the converter:
for ISO 3200 on +1.45 EV
for ISO 4000 on +1.00 EV
We insert frames into one PSD file, align (without scaling), we select an identical uniform section, without details and we watch dispersion of brightness on Std Dev, disconnecting a high layer and including it. Also we watch the signal level — Mean.
After we consider on a formula SNR = Mean/Std.Dev
So, for a fragment on ISO 3200 I received SNR=50.37/5.03=10.01, and for ISO 4000 SNR=51.36/4.69=10.95
The scene shot on ISO 3200 and tightened in the converter had more noise than the scene shot NAISO 4000, but it is less tightened in the converter.
It is the interesting moment. I take two more frames. ISO 400 and ISO 800. They are shooted with an identical exposure and a diaphragm.
The frame on ISO 800 will be without tightening of an expoztion in the converter, and ISO 400 we will tighten on 1 feet ideally to be made even to ISO 800.
for ISO 400 on +1.00 EV: SNR=30.89/2.29=13.48
for ISO 800 on +0.00 EV: SNR=30.68/1.94=15.81
Canon 1D X
There are two frames. Both are shooted on F8, ½ sec. Only one frame is on ISO 3200, and another is on ISO 6400.
Same lens: Carl Zeiss Makro-Planar 50/2 ZE.
We put correction of exposure (that it would havel visible noise and that the pictures matched on brightness ):
ISO 3200: +4.00 EV
ISO 6400: +3.00 EV
The camera Canon 1D X has less noisy matrix, than the Canon 5D Mark II. But this not so considerable advantage as the owner of such expensive camera would like.
I will add to logical council of the picture correct exposure the recommendation not to hope for the RAW converter and to raise ISO necessarily if there is no opportunity to shoot on more open diaphragm or with longer exposure.
Soon I hope to add this article to the second part where I will analyze ISO for both cameras Canon 1D X and the Canon 5D Mark II to a maximum of their opportunities.