Live Composite Mode / Live Time Mode

Started by micwag, September 25, 2016, 09:24:59 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

micwag

Hi,
I've seen a really cool feature on Olympus cameras:
You set the expoure time and the camera takes umlimited photos subsequently. These images are combined to one final picture. In Live composite mode, the brighter pixels of the newer images are added to the original photo, so the background doesn't get brighter, but new/moving lights get visible. In live time mode, the brightness of all images gets combined as in long exposure photographs.
In both modes, you can always see the progress after the last taken image.
Implementing this feature would require to calculate and display the result during the exposure of the next image. Would this be possible?

a1ex

Already requested here.

It can be done with the full-res silent picture mode, using these routines to combine the images.

The end result will have higher absolute noise (not as good in very low light), but also higher dynamic range. Math is left as an exercise to the reader.

Straight_Shooter

Quote from: a1ex on September 26, 2016, 12:57:33 AM
The end result will have higher absolute noise (not as good in very low light), but also higher dynamic range.

How so? I thought that when you combine many pictures into one, then you get less noise in the end result? Since the noise is randomly distributed in every picture, and you have as many data points for every single pixel as you have pictures, I thought that this would even out then.

Maybe I am mistaken here - quite possible. I am by no means a professional photographer. Rather an advanced amateur.

Also, I would totally kill to have these features from Olympus being available on Canon cameras! ;)
Or rather, I would pay 100 (one hundred) dollars to the person who implements Live Composite Mode and Live Time Mode in Magic Lantern. Any takers?

a1ex

Quote from: Straight_Shooter on January 15, 2017, 04:12:21 PM
How so? I thought that when you combine many pictures into one, then you get less noise in the end result?


octave:1> a = randn(10000,1);
octave:2> b = randn(10000,1);
octave:3> std(a+b)
ans =  1.4128


What increases is the SNR (combining 2 images gives signal = 2x, noise = 1.41x => SNR = 1.41x), which in turn results in higher DR (0.5 EV more, assuming Gaussian noise).

For N pictures, SNR and DR increase by log2(sqrt(N)) EV, but shadow detail will suffer by the same amount. For example, if you compare a single 32-second*) exposure with a sum of 32 1-second exposures, the latter is 2.5 EV worse in shadows, but 5 EV better in highlights (so DR increases by 2.5 EV, but it's no longer able to capture very deep shadows).

*) When you select a 30-second exposure, the camera actually does a 32-second one (try with a stopwatch).