| Author | Thread |
|
|
12/29/2004 11:38:45 PM · #1 |
Originally posted by faidoi:
Some cameras do it in camera now. |
Many cameras will do the dark-frame subtraction, but none that I am aware of will do multi-image averaging.
|
|
|
|
12/29/2004 11:23:20 PM · #2 |
Some cameras do it in camera now.
|
|
|
|
12/29/2004 11:20:54 PM · #3 |
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
|
|
|
|
12/29/2004 11:18:08 PM · #4 |
I'm surprised Laurielblack hasn't chimed in on this very subject, LOL.
|
|
|
|
12/29/2004 11:16:09 PM · #5 |
oh goodness gracious -- I was following along nicely then you mathmetician/mad scientists started adding in your pathagroean's therom...I'm lost :-D
|
|
|
|
12/29/2004 10:18:26 PM · #6 |
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.
|
|
|
|
12/29/2004 09:14:11 PM · #7 |
Originally posted by kirbic:
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. |
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?
|
|
|
|
12/29/2004 08:46:44 PM · #8 |
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.
|
|
|
|
12/29/2004 06:49:27 PM · #9 |
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.
|
|
|
|
12/29/2004 06:32:51 PM · #10 |
Right, detail. Look at it this way; when you use neatimage, it makes a "determination" whether a given pixel is noise or information. If it decides it is noise, it then interpolates information from neigboring pixels and overwrites the noise.
Now, take 5 separate, identical exposures in your camera. In each the noise, which is random, will be different. Imagine a program which analyzed each of these layers and, based on "majority: rules, assigned a "noise" or "information" tag to eaach pixel: this would be way more accurate, see? Id 3 or 4 of the images agreed a given pixel was information, it would get tagged that way. Imagine a composite image, cimnputer generated, of the highest "vote getters" among the pixels. This would be very noise free, and then neatimage could work on THAT if it wanted to and interpolate away the remaining, slight noise.
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.
Thus, more detail is preserved.
I use a 3rd party jpeg repair tool that I love, and it produces some fascinating effects if you dial it way up, blocking out monotone areas and leaving texture in complex areas. It works on the same principle
as neatimage, I believe.
Robt.
|
|
|
|
12/29/2004 06:30:22 PM · #11 |
Originally posted by GeneralE:
If you generate two (or more) versions from a RAW file with slightly different settings, can you affect the noise distribution? If so, you could DPC-legally stack those. |
Nope. You can affect the amplitude of the noise, but not it's spatial distribution (pattern). The overall idea is the ol' axiom that "entropy cannot decrease", that is, you can't pull information from noise, order from chaos. you need to add information, and that means a true second exposure. There truly is no free lunch.
Message edited by author 2004-12-29 18:31:18.
|
|
|
|
12/29/2004 06:30:14 PM · #12 |
Thanks JD this stuff is gold!!!
|
|
|
|
12/29/2004 06:27:42 PM · #13 |
| If you generate two (or more) versions from a RAW file with slightly different settings, can you affect the noise distribution? If so, you could DPC-legally stack those. |
|
|
|
12/29/2004 05:40:57 PM · #14 |
Originally posted by bledford:
Isn't this just the kind of stuff a tool like NeatImage already does for you? What advantage does this have other than keeping everything in photoshop (which can be done with the full version of most noise reduction plug-ins)? |
It preserves all detail. |
|
|
|
12/29/2004 05:18:47 PM · #15 |
Remember that in Photoshop CS Camera RAW has a noise reduction slider, you can adjust colour noise and luminance noise, it does a pretty good job that Noise Ninja or Neatimage can improve on.
|
|
|
|
12/29/2004 05:15:55 PM · #16 |
It's actually not hard at all to do. If you make no other adjustments, you can slap this baby together in 2-3 minutes. No tools required for assembly, LOL. It gets tougher when you're using adjustment layers. The easiest way is to clone off the 4-layer base image to a new file, flatten the layers, and treat that as your original for other effects.
(robt)
|
|
|
|
12/29/2004 05:14:30 PM · #17 |
I believe that there might be another method for reducing noise but I haven't investigated it yet (it has been on my to-do list). The idea is to manually set your camera to the exact same settings as a photograph who's noise you want to reduce and then take a photo with the lens cap on (maybe even wrap the camera with a towel to make sure there is no light entering the lens) then combine the photo with the main photo via layers. I'm not sure what layer blending properties to use. It is supposed to be the same technique as the dark frame subtraction method that many cameras utilize on slow exposures. What I like about this technique is that you could take your second photo at any time afterwards. The one x-factor that could pose a problem (if this technique works at all) is the temperature that you are shooting in. Different temperatures can effect the amount of noise in a photo (colder means less noise, warmer means more noise, relatively). It sounds interesting and if I try it out I will let you know whether it works or not. Maybe someone else has already tried this and can comment.
T
|
|
|
|
12/29/2004 04:59:13 PM · #18 |
Originally posted by kirbic:
Originally posted by rudidl:
For most photographs, no advantage at all. Use NeatImage.... |
Beg to differ... if you ever try it you will see that there is a huge difference. The only downside is that any motion at all will result in some funkiness. If motion blur is OK. like falc's waterfall, then this technique can provide some awesome results.
There's a second type of noise that's not addressed by this technique. It's known as fixed-pattern noise, and as you might guess, it has the same general pattern shot to shot. Because of this property, you can get rid of much of it by subtracting a "dark frame", essentially a pic taken with the same setings but the lens cap on. Serious astrophotographers do both dark frame subtraction and averaging. Also note that the 20D and I believe the D70 will do dark frame subtraction in-camera. |
Cool to know! Seems like an aweful lot of Photoshop work though. I guess I'll start worrying about that when I take the sort of pictures I can retake 5 times in a row *and* I'm worried about random pattern noise. LOL.
Message edited by author 2004-12-29 16:59:30. |
|
|
|
12/29/2004 04:59:12 PM · #19 |
Rudidl,
TRY it; it's not hard. Set up a shot of something static with a tripod, shoot it 4 times. Make one composite image layering all 4 images as above, make another staright image from one exposure, and then clone that and use neat image on it.
Blow 'em all up like 800% and compare. You'll be stunned. I'm not kidding.
(robt)
|
|
|
|
12/29/2004 04:56:30 PM · #20 |
Originally posted by kirbic:
Beg to differ... if you ever try it you will see that there is a huge difference. The only downside is that any motion at all will result in some funkiness. If motion blur is OK. like falc's waterfall, then this technique can provide some awesome results. |
Thanks! So we learn new things every day.
I have used both techniques with some astronomy photos, but have never tried it elsewhere. Cool, something new to play with the next few days! :-)
|
|
|
|
12/29/2004 04:07:58 PM · #21 |
Originally posted by rudidl:
Originally posted by bledford:
Man, I tried to read that thread, but my "inner monologue" started saying "blah, blah, blah...NeatImage" Isn't this just the kind of stuff a tool like NeatImage already does for you? What advantage does this have other than keeping everything in photoshop (which can be done with the full version of most noise reduction plug-ins)? |
For most photographs, no advantage at all. Use NeatImage.... |
Beg to differ... if you ever try it you will see that there is a huge difference. The only downside is that any motion at all will result in some funkiness. If motion blur is OK. like falc's waterfall, then this technique can provide some awesome results.
There's a second type of noise that's not addressed by this technique. It's known as fixed-pattern noise, and as you might guess, it has the same general pattern shot to shot. Because of this property, you can get rid of much of it by subtracting a "dark frame", essentially a pic taken with the same setings but the lens cap on. Serious astrophotographers do both dark frame subtraction and averaging. Also note that the 20D and I believe the D70 will do dark frame subtraction in-camera.
|
|
|
|
12/29/2004 03:17:52 PM · #22 |
Originally posted by Falc:
Its also quite a nice technique for capturing waterfalls. Each droplet gets forozen in a different place. This means that the areas where the droplets are close together become silky smooth, whilst the areas of wider spaced droplets become very much richer.
I used to do it with film cameras using multiple exposure and reducing the exposure values way down.
Have a play next time you go to the local waterfall, take your tripod and trip off 8 identical (almost) shots.. |
Just thought I'd put up an example of a stacked waterfall - this one is 10 images stacked as described above.
 |
|
|
|
12/29/2004 02:32:43 PM · #23 |
Originally posted by bledford:
Man, I tried to read that thread, but my "inner monologue" started saying "blah, blah, blah...NeatImage" Isn't this just the kind of stuff a tool like NeatImage already does for you? What advantage does this have other than keeping everything in photoshop (which can be done with the full version of most noise reduction plug-ins)? |
For most photographs, no advantage at all. Use NeatImage.
For things like Astronomy, big difference. That small dot that NeatImage thought was as a bit of noise was actually a new comet, heading straight for planet Earth. Too bad you missed it... :-)
|
|
|
|
12/29/2004 02:23:12 PM · #24 |
Originally posted by willem:
By the way,here is a link that shows the correct (?) percentages to be used. |
Man, I tried to read that thread, but my "inner monologue" started saying "blah, blah, blah...NeatImage" Isn't this just the kind of stuff a tool like NeatImage already does for you? What advantage does this have other than keeping everything in photoshop (which can be done with the full version of most noise reduction plug-ins)? |
|
|
|
12/29/2004 02:15:53 PM · #25 |
Its also quite a nice technique for capturing waterfalls. Each droplet gets forozen in a different place. This means that the areas where the droplets are close together become silky smooth, whilst the areas of wider spaced droplets become very much richer.
I used to do it with film cameras using multiple exposure and reducing the exposure values way down.
Have a play next time you go to the local waterfall, take your tripod and trip off 8 identical (almost) shots.. |
|
