There would be tested 2 cameras:
-Canon 5D mark II
-Canon 1D X
The second camera is the newest generation of top cameras from Canon, especially is interesting to compare it to the previous generation of cameras Canon 5D Mark II. Whether you will have a benefit on noise on the camera at the cost of 7700 USD?
In Adobe Photoshop there is a Std.Dev value, in the tab Histogram which defines dispersion of values.
If on the tab to select from a dropdown menu Luminosity, we will measure in the channel of brightness that allows us to measure the most unpleasant brightness noise. There is still a color noise, but in this case I am not sure yet that its quantity can be so simply evaluated.
Except everything the same tab shows us Mean value (average). I.e. it is average brightness for all provided values, so it can be a certain signal of a digital signal if to analyze the uniform plane on a photo.
So, we have a signal level (Mean) and the maximum deviation (Std.Dev) from it. From them we already receive required SNR (Signal to Noise Ratio, Signal/noise).
It is important to measure this ratio as Std.Dev value in case of even a little different brightness of a picture will give us strongly different absolute value of noise. And we need value of noise concerning the signal level. Thus we will exclude the slightest errors on brightness.
SNR = Mean/Std.Dev
Gann, Robert, Ph.D is offered “. Desktop Scanners: Image Quality Evaluation”, Hewlett Packard.
Noise on the Canon 5D Mark II
Reference frame on ISO 100. Noise reduction is disconnected, F8. Lens Carl Zeiss Makro-Planar 50/2 ZE.
On light areas (bricklaying) there is no noise, and in window openings appeared scarcely noticeable noise.
ISO 100, SNR=74.05/4.45=16.64
ISO 200, SNR=73.28/4.59=15.96
Than above SNR, it is respectively better (less noise).
As you can see, noise in window openings which it seems not such and dark, are already quite visible. I didn’t draw out exposure in the converter here (if suddenly you have such thought).
ISO 400, SNR=74.85/4.85=15.43
Here not absolutely correctly because I saw a darkened picture with ISO 800 on 1 step. It partially corrected noise. If to light and blur some areas, then noise will be less. A lack of this method — if light some areas, then it will be impossible to correct the beaten-out areas (on 5D of mark II I have0.5-0.7 EV relights in a RAW format will be quietly reanimated in the converter).
Generally the correctness of such comparing will always be disputable. My method is the reduction of an exposure and increase of ISO. It is logical reception for all who remove moving lenses in the twilight and who need really high ISO for operation. It both photographers of weddings and sporting photographers of all remaining areas (office parties, actions, etc.)
The second option would be in saving an exposure of a picture with the bigger ISO in the converter.
I think that this option is wrong because it is quite strange to shoot with overexposure and then to try to reanimate a picture. The “beaten-out” areas will often meet. Also it is strange to shoot with a lighting in the converter. I is hard to tell something about shadows and suites on picure (as well as according to the histogram), and the result raises big doubts. Here experiment says that there are less noise if to shoot on adequate ISO and once again not to touch the converter. At the least modern ACR, and to it is more than trust, than to remaining, crude converters.
ISO 800, SNR=75.72/4.66=16.24
As you can see, mine influenced reduction of noise.
ISO 1600, SNR=72.70/4.76=15.27
ISO 2500, SNR=72.71/5.48=13.26
ISO 12800, SNR=73.89/6.76=10.93
ISO 25600, SNR=66.92/7.74=8.64
ISO 25600 is completely dummy. There is a big loss of details both in light, and in dark areas. Then only to reduce such photo for use by an avatar.
I will repeat that for the camera Canon 5D Mark II I would stop on ISO 800 when light areas are almost completely free from noise, average-gray it is possible to correct noise reduction with thin settings without loss of fine details, and dark to leave as it is.
On ISO 1600 you already begin to lose fine details in average-gray. Any noise reduction won’t return them to you. I.e., for example, it ISO doesn’t suit for a landscape at all.
Canon 1D X
ISO 100, SNR=57.76/4.30=13.43
ISO 200, SNR=64.12/4.57=14.03
I selected area in a window opening and I measured it programmatically there is a difference after all.
ISO 400, SNR=63.22/4.54=13.92
However, optimum for this camera is ISO 200, but not 100.
Canon 1D X: SNR=62.59/4.40=14.225
Canon 5D mark II: SNR=75.63/4.65=16.26 (-1 EV)
Well, wins against a picture with the Canon 5D Mark II which was overexposed and darkened on 1 step.
ISO 3200, SNR=61.54/6.96=8.84 (+1 EV).
Result of tightening in the converter is noise.
ISO 5000, SNR=53.61/5.52=9.71
ISO 12800, SNR=70.32/9.29=7.56
ISO 51200, SNR=60.53/18.49=3.27
ISO 102400, SNR=58.34/24.30=2.40
First cameras according to this table can be compared only to ISO 400 inclusively because only to ISO 400 I used an identical exposure.
The result surprised me. It turned out that on the big dark areas of Canon 1D X benefited on noise, and here on lighter sections, as window opening on these fragments it loses on noise. I have no explanation yet, one guesses. Perhaps characteristics of a sheaf the matrix-ATP-converter — are non-linear. Then it strongly complicates our tests for noise.
It is clear, that the picture rustles differently in dark and lighter areas. And also it is clear that it is more important to pull out details from dark areas as in light the picture by determination rustles less.
But why newer camera rustles less in shadows more in lighter areas is still unclear.
I can issue all RAW files. I rechecked results and even counted everything numerically, as you can see. Both visually it is more noise and numerically it is estimated as the bigger.
In spite of the fact that I made change of an exposure it is impossible to compare noise, iso 400, there was an opportunity to make other test.
There is such technique of ETTR (Exposure to the right) which recommends to offset all histogram to the right. Up to relight on 1 feet. These 1 feet then it is subtracted in the converter for receiving the correct exposure, but we have much less noise, than directly shooting with normal one (according to the camera histogram) exposures. It is well illustrated by blue arrow on the histogram. As the technique of ETTR was used the ratio of the useful signal does a saltus up.
Also pay attention to ISO 3200. There the single case of “extraction” via the converter and a ratio the useful signal / noise falls directly, confirming results of the previous test.
I will go further to analyze pictures from both cameras on ISO 100 — ISO 400 (where all parameters of shooting are identical) in different areas (bright, average, dark) to collect as much as possible information and to draw a final conclusion on characteristics of noise to the Canon 5D Mark II and Canon 1D X.
I wish to adopt ETTR for shooting of a landscape.