Re: What I learned from Gollywop -- and what I wonder

1

Macx wrote:

Anders W wrote:

Macx wrote:

olliess wrote:

Macx wrote:

Stacking images should give you at least as good noise performance in the shadows as bracketing.

If you only stack (instead of bracketing), won't your shadows potentially take multiple "hits" from pattern noise, black clipping (if your camera does it), etc.?

Yes, that is a good point, and it's worth testing how much this adding up of "per-shot noise" will mean for the final image. There might very well be a break-even point. For the E-M5 the sensor noise is fairly low, so I would suppose that the perceived noise (signal to noise ratio) in the end image would still be lower, even with multiple exposures, because the read noise is fairly small compared to the shot noise caused by (lack of) exposure.

No, I think stacking only, without any exposure bracketing, is perfectly fine and it is sometimes the best you can do. It is merely less efficient than stacking with exposure bracketing when that's possible.

The problem with the stacking-only strategy, is that you need a whole lot of exposures to get anywhere. With stacking only, without exposure bracketing, DR grows in proportion to the square root of the number of exposures. So to get one EV more of DR you need four, not two shots. To get two EV more DR, you need 16 shots, to get three EV you need 64, and so on.

Are you sure about this? I would have thought that doubling the exposure would allow for one more EV, and it would be four exposures for two EV and so on? And while the many exposures is definitely the impractical bit, both during capture and when processing, but beside that I think I think it's fairer to look at the total sum of exposure time from the two methods.

Yes, I am sure. The standard "engineering" definition of DR is

DR = S/R

where S is the max signal and R the read noise. S grows in direct proportion to the number of exposures. However, R grows as well, more specifically in direction proportion to the square root of the number of exposures.

Consider an example with four merged shots. S will then be four times as large as in a single shot. But R will be sqrt(4r^2) = 2r, where r is the read noise in a single exposure.

You find the mathematical underpinnings here if you are interested:

http://en.wikipedia.org/wiki/Sum_of_normally_distributed_random_variables