Noise Reduction at the Source

[brian_m2]

I've seen some hints on other threads about techniques to use to

reduce high iso noise when taking the photo in RAW - the one that

comes to mind is to make sure you don't underexpose because using the

PS CS RAW converter will increase/accentuate noise in the shadows.

I'm using very good noise reduction software but it wasn't able to

take care of all of the noise in an iso 1600 photo from my d70. Shot

with available low tunsten lighting in front of a Christmas tree;

corrected the lighting and raised exposure by 1.5 stops in CS RAW

converter. Terrible noise.

 

Besides correct exposure (why my d70 underexposed like that is

another question), what else can I do in technique to reduce high iso

noise?

 

And, is there something more sophisticated than just running it

through noise reduction software that I can do in PS CS to reduce

noise without artifacting?

 

Thanks.

[beauh44]

Hi Brian, The only thing I can think of is temperature. I've heard that digital sensor noise decreases in cold temperatures. I've never tried it, but I've wanted to test my Canon 10 or 20D by putting it in the freezer for a few hours (without the battery!) to see if it does anything at all. I suspect I might have to have colder temperatures than that though. Perhaps you could give it a shot and let us know if you can tell the difference between an "unfrozen" image and a "frozen" one. Good luck!

[charlie_vigue]

Beware condensation. I've shot some (on a telescope, in the Oregon High Desert) in very cold conditions and it did seem to reduce noise. Also made my toes cold. I was doing astrophotography in winter, temp was between -10 and 0 F, no wind.

 

Bringing it back indoors will generally make it form condensation until it equalizes again.

[tom_swanson]

Use a better RAW converter. ACR generates noisier shadows than any other RAW converter I've used. (All for Canon DSLRs.)

 

Try the Phase One C1 LE demo.

[jeff medkeff anchorage, a]

<p>The previous poster is correct that a certain type of noise will be reduced with lower sensor temperature. You are going to experience three dominant types of noise with a digital camera:

 

<ul>

<li>Photon noise aka Poisson noise, which refers to natural variation in the incident photon flux or, put a bit more simply, it refers to the fact that photons don't always arrive at the sensor in a nice, regular, orderly manner. There is nothing you can do about this type of noise except increase the amount of time the shutter is open.

<li>Dark noise results from the variation of thermally generated electrons in the sensor itself. (The closely-related term <i>dark current</i> refers to the rate of generation of thermal electrons at a given sensor temperature.) Dark noise gets worse at higher temperatures and with longer integration times (integration time is very closely related to exposure time in a DSLR but is not necessarily 1:1).

<li>Readout noise results from quantifying the signal on the sensor. Reading out a sensor involves measuring the charge at each photosite on the sensor; readout noise is a result of such measurements being imperfect. Most of the readout noise in DSLRs results from the preamplifier. Readout noise is the same regardless of exposure time and varies only a small amount as sensor temperature changes.

</ul>

 

<p>By increasing the ISO setting on your camera, you are increasing the amount of amplification of the signal read out from the photosites on the chip. If you were shooting at hand-held shutter speeds, the dark noise component of your noisy image will be close to neglegible. So your question probably reduces to, 'what can be done to reduce the visible effects of read noise?'

 

<p>I would shoot a bias frame and subtract it from the original image. My workflow for this would be:

 

<ol>

<li>Take your camera, and come up with a way to prevent any light entering the lens. Take a photo with the light-preventer in place. Make sure the camera is set to the same ISO and shutter time setting as the noisy photo you described. We will (with a little generosity) call the resulting image the "bias frame."

 

<li>Go back to your original RAW image, and convert it to TIFF, setting white balance and, if you must, exposure compensation only. Do no further processing at this time; ESPECIALLY don't be tempted to alter levels if your RAW converter allows this.

 

<li>Apply the same RAW conversion settings to the bias frame and convert it to TIFF. White balance and exposure compensation settings in the RAW conversion are fundamentally critical and you must insure these are exactly the same with this conversion as they were on your image conversion; otherwise this won't work at all.

 

<li>Load both your image and the bias frame in Photoshop. Go to the Image menu, select Calculations. For Source 1 select your image. For Source 2 select your bias frame. For Blending select "Subtract," opacity 100%, offset zero, scale one, leave "Mask" unchecked. Send the "Result" to a new document so you can compare before and after.

</ol>

 

<p>At this point the result could go two ways. If the original image has decent signal, you will see a dramatic reduction of noise in the shot. You can declare victory. You can attack any residuum of noise with Noise Ninja or the like. Whatever.

 

<p>But it is possible that photon noise is so high in an underexposed image that subtracting a bias frame is sort of futile. You could still end up with an unacceptably noisy image. You could also achieve failure due to mistakes (or bugs) in RAW conversion. You might find that subtraction of the bias frame leaves the image tonally very flat. If so, you can try the bias subtraction again but with 75% opacity, etc, but don't expect miracles - at this point you are dealing with so few actual bits of data that at some point you may have to admit the shot is hopelessly underexposed.

<p>Good luck.

[charlie_vigue]

In my very cold shooting the shutter times were on the order of minutes, so that's probably why the ambient temperature made a visible difference.

[jeff medkeff anchorage, a]

<p>I've written a first draft of an <a href="http://medkeff.com/jeff/photo/noise/">article about digital SLRs and dark/bias frame subtraction</a>, which expands and corrects some of the stuff I previously posted in this discussion thread. I intend to offer this article for use on photo.net, if I can figure out who to make the offer to. At any rate, I throw this out in case anyone is still interested.