This is a change of subject since the original thread ... it’s often better to just create a new thread.
Anyway, in the image you posted, I am not seeing any noise ... which makes me wonder what you’re seeing. I am specifically looking at the sock in the lower left corer (since that’s basically a black area). If the image is noisy... that area should reveal it. But I’m not seeing it in this image (possibly due to resizing).
Re-sizing an image to a smaller size involves reducing more pixels into fewer pixels. E.g. if you make an image half as wide and half as high, then that means each 2x2 cluster of pixels (4 pixels) will be reduced to just one pixel. If ONE of those 4 pixels was excessively noisy it will get averaged out by the other three non-noisy pixels and this naturally reduces the noise.
To truly evaluate the noise, we’d need to see a full-size version of the image (preferably the RAW “.CR2” file since JPEG images are typically auto-processed to reduce noise.
So a bit about noise...
ALL cameras have noise. But the reason you might see one image and think it has no noise and see another image and think it has a lot of noise has to do with something called the “Signal to Noise Ratio” (SNR).
Basically the noise level remains somewhat constant... but it’s the signal level that varies. If you don’t collect enough “signal” (good data) then the ratio is worse and you perceive a noisy image.
Exposure is based on ONLY shutter time and aperture size. ISO is not part of “exposure” because ISO is an amplification of the data which occurs only after the shutter closes and the image is captured. At that point any amplification is amplifying everything ... good signal and noise alike.
This is why people feel shots taken at low ISO don’t have noise and high ISO do have noise. The reality is that the amount of noise is the same. But in the low ISO shot, you either had more available light (good signal was arriving in abundance) or you allowed a larger aperture or longer exposure time until the camera collected enough signal... such that there was no reason to amplify the data. That results in a high amount of signal relative to the noise (a good SNR) so the noise isn’t amplified and you don’t perceive it (even though it really is there ... every digital sensor has noise.)
If you were to leave the lens cap on and take the fastest possible exposure at base ISO (ISO 100), then inspect the pixels in that “black” photo (since you would have guaranteed no light could enter the camera), you’d find the pixel values aren’t actually zeros... there very low values just fractionally above 0. This is because in order to operate the sensor, it has to be powered up (voltage must be applied) and there’s a minimum charge that can be read out from each pixel as a result. This is called the “bias” level. Also, you’d notice that the pixels wont all have the same value.
When the camera performs a read-out of the sensor, it collects “read noise”, there is also “thermal noise”, there’s even quantum noise (which can’t be eliminated ... it’s in the nature of the universe).
There are many kinds of noise that collect... but in order to not notice the noise you need to collect enough “signal” that the signal (good data) overwhelms the noise so that the noise simply isn’t noticed.
Enough on “why” you have noise (you can find loads of technical articles on noise if you want to know more). Once you have It, you can still do things in post-processing to reduce it.
”noise” is basically a pixel that reads a slightly higher value than it’s neighbors. This really tends to be noticed in dark shadowy areas... and also in smooth non-contrasty areas. It tends to not be noticed in bright areas and is more difficult to pick out in areas with lots of details (it can get camofauged in the detail).
The process of de-noising involves sampling neighboring pixels and then averaging the colors. But in areas of an image with high detail, this can result in the image starting to look soft. BTW, sharpening is the opposite... this involves finding subtle differences in contrast... and amplifying those differences. Since a “noisy” pixel is already different in contrast, when you “sharpen” an image you will increase the perceived noise.
So to deal with this, use tools that let you selectively sharpen only the areas you care about (e.g. eyes for example) and you can selectively de-noise the smooth areas.
There are de-noising tools that will deal with noise by tonal frequency (since noise tends to be stronger in shadow areas). But you can also build edge-masks so that you avoid de-noising edges that need to remain sharp... then invert the mask to sharpen only those edges without sharpening noise in the flat areas.
Tim Campbell
5D III, 5D IV, 60Da
