Black level unchanged after long exposure noise reduction

Question

I took 2 raw photos with the lens cap on: One with Long Exposure Noise Reduction turned on, one with LENR turned off.

Checking the tags in the Raw file, BOTH report black levels of [200, 200, 200, 200] (One for each pixel in the Bayer pattern, I suppose). Total range is 4095 counts (12-bit)

I would expect this to be the case for the LENR=OFF photo, but not for the LENR=ON, since it subtracts a dark frame from the (also dark) photo, removing any offsets. Why does the black level stay the same?

EDIT: Exposure was 10 seconds in both cases. I have also added the histogram of both images, which might help. Granted, at 10 seconds, even with LENR turned off a second exposure is taken of ~3 seconds, but I'm not sure what happens there. The biggest question for me now is why the black level remains 200 after frame subtraction.

Answer 1

Noise reduction is concerned with maximizing the Signal-to-Noise Ratio (SNR) by differentiating between actual signal and signal created by noise. With a lens cap on the camera there is no signal in terms of electrons excited by photons belonging to light entering the camera that are striking the imaging sensor. It is all noise. How the camera acts under such conditions may not be indicative of how the camera acts under the conditions for which the LENR system was designed: reducing predictable noise from an image with actual signal in it.

Answer 2

There are many sources of noise in the CMOS imaging process. Fixed-pattern noise (FPN) refers to image noise caused by nonuniformity in the sensor. For example, differences in manufacturing might cause some pixels to be slightly more sensitive to light. Or interference from other circuitry might cause pixels in some areas to report a different value temporarily.

Here is an example. This is what long exposure noise reduction is designed to subtract out.

However, fixed-pattern noise is not the only source of noise. Even for a dark shot, the sensor will still report photons due to thermal noise and dark current noise. Unlike FPN, the effects of these types of noise will change from shot to shot.

I'd guess that your "10 LENR" photo has less fixed-pattern noise than the one without LENR. However, because of the other types of noise, you're subtracting a random number from another random number at each pixel, ending up with a random number. That's why the histograms from both images look similar.

As for the black level reported in the RAW file, I found this paper which suggests they are constants for the given model of camera:

Understanding What is stored in a Canon RAW .CR2 file, How and Why (Section 5.2 Black substraction)