Shh....it's a secret -> Noise Reduction.

Nikon AF-S Nikkor 28-300mm f3.5-5.6G ED VR

12/29/2004 06:49:27 PM · #26

Originally posted by bear_music:

...Now, that's not what we're doing, of course. To my knowledge no such program exists at the photoshop level. However, by layering together the various exposures at varying degrees of transparency, we are "overwriting" a LOT of noise with "real" information instead of interpolations of what ought to be there...

Actually, I'd maintiain tht it is exactly what we're doing. By overlaying the images with the appropriate opacities, we are in faact giving each pixel a value based on a the mathematical averages of the R,G and B values for all of the layers. AFAIK, there is no more efficient algorithm available for reduction of random noise.
BTW, for the techies, the noise level is reduced as 1/n^(1/2) (one over square root of the number of overlayed frames). So for instance, 2 frames yields 1/2^(1/2)=.7071, so noise is reduced to 70% of original level. For 4 frames, it's 50%, 8 frames 25% and so on.

Arcanist

Nikon D810

12/29/2004 08:46:44 PM · #27

Originally posted by kirbic:

...BTW, for the techies, the noise level is reduced as 1/n^(1/2) (one over square root of the number of overlayed frames). So for instance, 2 frames yields 1/2^(1/2)=.7071, so noise is reduced to 70% of original level. For 4 frames, it's 50%, 8 frames 25% and so on.

I was hoping someone would get mathematical on us here. The 'alternate' percentage calculation that was referred earlier intrigued me but I am quite sure the resulting dialogue neglected the complexities of opacity in the other layers and the effects of a 100% hot vs 75% hot modified, compared to 50% or 25%.

The simplest conundrum I could come up with to support this ACE Photoshop Tutorial's article is that at 100%(background) hot, a 75%correct+50%correct+25%correct =150% correction of color (using 4 separate exposures) but we still have to take into consideration that at L1, 25% of the hot pixel is bleeding through to only a 50% Strength of color correction. ACK!!!!! Here I go again, I'm a philosopher, Jim, not a mathemetician! I might as well be separating my molecules and trying to recombine them.

nicklevy

Canon EOS-350D Rebel XT

Tamron SP AF 28-75mm f/2.8 XR Di for Canon

12/29/2004 09:14:11 PM · #28

Originally posted by kirbic:

So would it mean using this technique with 8 different at ISO800 we would be left with the same amount of noise as if we were shooting at ISO100?

Nikon AF Nikkor 50mm f/1.8D

12/29/2004 10:18:26 PM · #29

Originally posted by nicklevy:

So would it mean using this technique with 8 different at ISO800 we would be left with the same amount of noise as if we were shooting at ISO100?

Interesting question. The answer is, it depends on how the noise in the system increases with increasing ISO. Let's do an empirical exercise. For 8 exposures, noise will decrease to 1/8^(1/2), or 1/2.8, or about 35% of the original noise.
Most DSLRs have luminance noise of approx. 2.5 to 3.0 at ISO 800, and 0.9 to 1.15 at ISO 100. So is the ratio of 2.75:1.05 approximately equal to 2.8:1? It actually is, approximately, so yes, it does mean that using 8 exposures, the noise at ISO 800 will look pretty much what you normally see at ISO 100!
Not only that, but the "rule" holds up well through ISO 3200! My suspicion is that the mathematical relationship governing the increase in noise with increasing ISO is similar in form to the equation governing noise reduction using averaging. If I thought about it long enough, I could prolly dig out the reason, but by that time even I'd have a headache ;)
BTW, my source for DSLR noise data was DPReview's review of the Canon EOS 20D.

deapee

Kodak V803

12/29/2004 11:16:09 PM · #30

oh goodness gracious -- I was following along nicely then you mathmetician/mad scientists started adding in your pathagroean's therom...I'm lost :-D

Sigma 105mm f/2.8 EX DG Macro for Canon

12/29/2004 11:18:08 PM · #31

I'm surprised Laurielblack hasn't chimed in on this very subject, LOL.

longlivenyhc

Canon EOS-300D Rebel

12/29/2004 11:20:54 PM · #32

Originally posted by deapee:

oh goodness gracious -- I was following along nicely then you mathmetician/mad scientists started adding in your pathagroean's therom...I'm lost :-D

im with you lol

faidoi

Pentax K10D

Pentax DA 18-55mm f/3.5-5.6 AF

12/29/2004 11:23:20 PM · #33

Some cameras do it in camera now.