I have my in-camera noise reduction set to On in my 5D3. Is that what most folks do, or do you shoot without and deal with noise in post? I imagine there's no single answer to this!
Paul - I have mine set off (in my Olympus EM-5) simply because I want to control it in post. It's possible - especially on newer cameras - that the in camera reduction might be superior to what you can do, or certainly less effort. I'd suspect it very much depends on the camera/maker and maybe some tests would show you one way or the other.
I can't comment on the 5D3, but I know I'll be testing my EM-5 eventually (it's only a few weeks old).
I've tried it, but I don't see any difference in the raw file. Likewise for long exposure nr.
I don't use DPP, but I tried it to see if it has a function that invokes in-camera/long exposure nr. Couldn't find anything.
I corresponded on nr with Dave Kingham, who has spent countless hours studying night photography. Dave says the best way by far is blending a lower iso fg shot with the sky shot. Not hard to do, but still not as good as doing it in one shot.
Never use in camera noise reduction on my 5D Mark III. Plain truth is, not needed as this amazing camera gives me good noise results without it. For some cameras turning it on will also slow down your shooting since the camera must make two identical exposures - one the subject and another as if the lens cap was on so as to map the noise. Not sure if this is the case with the 5D3 since I don't use it but either way, I just don't want all this hocus pocus going. If doing long exposure night work, it would quickly become a pain I-T-A. Finally, as with David, I prefer to control these adjustments if needed in my post production.
I also have mine turned off. I suspect that this feature is more for eliminating hot or dead pixels via mapping a dark framerather than reducing any noise.
I set mine off Paul. It is exactly what John said, it is to take a completely dark frame to eliminate hot pixels. If you need to do this because you've got a hot pixel I would recommend taking a dark frame yourself and running the two frames through something like PixelFixer (http://pixelfixer.org/ ) It's great for astrophotography when you can get hot pixels from the long exposures or actually for any long exposures where the sensor can heat up.
The single most effective means of reducing noise is to reduce its capture!
How? Shoot RAW and utilize all possible DR existing to the right of the JPEG-adjusted histogram in your camera. This technique is termed Expose Beyond the Right (EBTR). How much DR is there? gotta experiment to find out...EACH CAMERA! I've tested more than one camera of a single model and found no two that shared the same amount of Extra RAW-Accesible Dynamic Range. One DSLR had 2/3 stop of ERADR; another had one and 2/3 stops, and a third had two and 1/3 stops! Gotta test every one.
To use that ERADR, set up ETTR exposure and add that camera's predetermined ERADR and make the exposure. With that exposure you may, for the first time, have actually used the appropriate exposure by which to capture the maximum of RAW image data quality...data with the highest Signal-to-Noise Ratio and thereby the lowest possible to practically no noise. How? By capturing all your image data as far away as possible from the mother lode of Noise way back there in the left half of the old JPEG-adjusted histogram.
But understand that that EBTR is not appropriate for all images. It is most conveniently employed for landscapes, portraits, and relatively stable subjects under unchanging illumination. Sports and active wildlife that move from sunlight to shade and back....not so good.
you'll have to get used to the washed out scene with blown highlights that will be displayed by that JPEG-adjusted thumbnail, but when you tonally normallize in ACR it's a mind blower to watch a perfectly exposed image revealed as you slide the exposure slider to the left!
And when you do start using it you'll be absolutely amazed at the scant to almost imperceptible amount of noise you'll capture.
"I have my in-camera noise reduction set to On in my 5D3. Is that what most folks do, or do you shoot without and deal with noise in post? I imagine there's no single answer to this!"
Perhaps no single answer, but probably some better answers than others.
1st, I assume you mean for low light photography. In low light photography, remember that ISO does not change sensitivity, and noise is dominated by that from photons (Poisson statistics). The more light you capture, the better the signal-to-noise ratio. Boosting ISO does not change the amount of light captured; only exposure time and/or lens aperture area change the light. Going above ISO 1600 on a 5D3 and most other cameras gives no advantage in noise (S/N), but is a disadvantage in reduced dynamic range as ISO is increased. The web page referenced above by David Kingham says use ISO 6400 for night photography. That has the effect of reduced dynamic range, more saturated stars, thus losing star colors and no benefit to S/N.
On a 5D3 (and other cameras) for low light work, one simply needs an ISO to boost the signals from the sensor high enough that pattern noise is not an issue. See the pattern noise in the sensor analysis at: http://www.clarkvision.com/articles/evaluation-canon-5diii/
and look at Tables 2a, 2b. By ISO 1600 the pattern noise is small enough to be a non issue. (This was not the case with the 5D2.)
Next, before a low light imaging session, do a forced sensor clean. This is a menu selection to clean now, not the manual clean, but the one that shakes the sensor and mirror to shake loose any dust. What this does is also update the hot/dead/stuck pixel list that gets recorded in the raw file. Then a good raw converter will read that list and delete those pixels from the converted raw file.
The 5D3, as most modern digital cameras, have on sensor dark current suppression. This greatly reduces pattern noise (both low and higher spatial frequencies) so that the dark frames are very clean. There is no need for in camera dark frame subtraction (known as long exposure noise reduction).
With the hot/dead/stuck pixels map. a lot of annoying noise is removed, and working at ISO 1600 gives the maximum dynamic range and low pattern noise for the faintest light.
If, after all this, your exposure still results in more noise than you want, it is best to suppress it in post processing where you have more control.
For more info on night photography see: http://www.clarkvision.com/articles/nightscapes/
David B./Lance/John R./Preston/John M./Hali, you have unanimously convinced me, off it goes! I'll run some twilight test shots this week and run it through the paces in post. Thanks!
Keith, right, good point, I meant long exposure. But, I'll run the same tests at high ISO with that filter off too, and see how it cleans up. I'm at 100 for landscape stuff, but definitely 1600+ for Milky Way and low-light portrait work.
David Graham, I will definitely give that a whirl. I've noticed that my 5D3 can pull detail out of its left hockey skate when I wouldn't think it was there in an overexposed image, much more so that my old 5D2. I just thought it was Canon improving things and I'm sure it is, but maybe I have some extra wiggle room on the right in this particular camera. Thanks for sharing that.
To add to others' points, Paul, in testing I've noted some loss of feather detail in avian photography with high ISO noise reduction turned on in my Canon 1D-X. Since feather detail is what I am after, it's counter-indicated.
Me being something of a newbie to using a DSLR, would you please outline the steps I'd need to take in order to "Expose Beyond the Right (EBTR)." I'm a little confused as to the process.
ETTR, EBTR are simply methods for maximizing the exposure. In raw data the blue and red channels typically have lower signals than the green. With different white balance settings, the red and blue data values are scaled up (multiplied). Thus the green channel typically saturates first as exposure time increases. But when one channel saturates, the other two may not, thus, it is possible for a raw converter to estimate those areas that saturated, and thus recover the information, at least approximately. This effectively gives headroom to a digital exposure from the metered level. For example, here is an exposure latitude test of a Canon 1D Mark II:
http://www.clarkvision.com/articles/exposure_latitude-1/
You can seen that the meter +2 stops still gives good color rendition for all colors in the test chart.
The problem with EBTR is one is relying on the meters, and exposure meters can be fooled in many ways, leading to many ways to saturate the image.
A better way to determine exposure is to monitor the color histogram (not just the luminance histogram). Do a series of exposures and monitor the RGB histogram. When the signal hits the right edge on the histogram, is the raw data actually saturated in that channel? If not how far beyond the right edge can one push it before the raw data are really saturated? If the camera histogram is accurate, going even 1/3 stop over right the edge of the histogram results in saturation in that channel (which again may be recoverable by raw converter reconstruction of the signal).
But in low light photography, as I discussed in my earlier post here, one need not go above ISO 1600 on most cameras. At that ISO, one is digitizing so much detail that even if one underexposes a stop then boosts the signal in post processing there is no loss. Thus there is no need to worry about ETTR or EBTR. To maximize exposure on the subject is not a function of ISO, only exposure time and lens aperture.
The only time I use ISOs higher than 1600 is in low light situations where I need to see the results on the back of the camera just after the image is obtained. For example, I use ISO 6400 for focusing on stars in a night shot, then drop back to 1600 for the real exposure to give more dynamic range.
Discussion
13 repliesI can't comment on the 5D3, but I know I'll be testing my EM-5 eventually (it's only a few weeks old).
I don't use DPP, but I tried it to see if it has a function that invokes in-camera/long exposure nr. Couldn't find anything.
I corresponded on nr with Dave Kingham, who has spent countless hours studying night photography. Dave says the best way by far is blending a lower iso fg shot with the sky shot. Not hard to do, but still not as good as doing it in one shot.
Long Exposure NR: Off
High ISO NR: Off
I prefer to perform NR in post processing using the Neat Image plugin for CS6. I like having the finer control over NR, and Neat Image does this well.
--P
How? Shoot RAW and utilize all possible DR existing to the right of the JPEG-adjusted histogram in your camera. This technique is termed Expose Beyond the Right (EBTR). How much DR is there? gotta experiment to find out...EACH CAMERA! I've tested more than one camera of a single model and found no two that shared the same amount of Extra RAW-Accesible Dynamic Range. One DSLR had 2/3 stop of ERADR; another had one and 2/3 stops, and a third had two and 1/3 stops! Gotta test every one.
To use that ERADR, set up ETTR exposure and add that camera's predetermined ERADR and make the exposure. With that exposure you may, for the first time, have actually used the appropriate exposure by which to capture the maximum of RAW image data quality...data with the highest Signal-to-Noise Ratio and thereby the lowest possible to practically no noise. How? By capturing all your image data as far away as possible from the mother lode of Noise way back there in the left half of the old JPEG-adjusted histogram.
But understand that that EBTR is not appropriate for all images. It is most conveniently employed for landscapes, portraits, and relatively stable subjects under unchanging illumination. Sports and active wildlife that move from sunlight to shade and back....not so good.
you'll have to get used to the washed out scene with blown highlights that will be displayed by that JPEG-adjusted thumbnail, but when you tonally normallize in ACR it's a mind blower to watch a perfectly exposed image revealed as you slide the exposure slider to the left!
And when you do start using it you'll be absolutely amazed at the scant to almost imperceptible amount of noise you'll capture.
Try it! You'll like it.
Dave Graham
East River, SD.
Perhaps no single answer, but probably some better answers than others.
1st, I assume you mean for low light photography. In low light photography, remember that ISO does not change sensitivity, and noise is dominated by that from photons (Poisson statistics). The more light you capture, the better the signal-to-noise ratio. Boosting ISO does not change the amount of light captured; only exposure time and/or lens aperture area change the light. Going above ISO 1600 on a 5D3 and most other cameras gives no advantage in noise (S/N), but is a disadvantage in reduced dynamic range as ISO is increased. The web page referenced above by David Kingham says use ISO 6400 for night photography. That has the effect of reduced dynamic range, more saturated stars, thus losing star colors and no benefit to S/N.
On a 5D3 (and other cameras) for low light work, one simply needs an ISO to boost the signals from the sensor high enough that pattern noise is not an issue. See the pattern noise in the sensor analysis at: http://www.clarkvision.com/articles/evaluation-canon-5diii/
and look at Tables 2a, 2b. By ISO 1600 the pattern noise is small enough to be a non issue. (This was not the case with the 5D2.)
Next, before a low light imaging session, do a forced sensor clean. This is a menu selection to clean now, not the manual clean, but the one that shakes the sensor and mirror to shake loose any dust. What this does is also update the hot/dead/stuck pixel list that gets recorded in the raw file. Then a good raw converter will read that list and delete those pixels from the converted raw file.
The 5D3, as most modern digital cameras, have on sensor dark current suppression. This greatly reduces pattern noise (both low and higher spatial frequencies) so that the dark frames are very clean. There is no need for in camera dark frame subtraction (known as long exposure noise reduction).
With the hot/dead/stuck pixels map. a lot of annoying noise is removed, and working at ISO 1600 gives the maximum dynamic range and low pattern noise for the faintest light.
If, after all this, your exposure still results in more noise than you want, it is best to suppress it in post processing where you have more control.
For more info on night photography see: http://www.clarkvision.com/articles/nightscapes/
Roger
Keith, right, good point, I meant long exposure. But, I'll run the same tests at high ISO with that filter off too, and see how it cleans up. I'm at 100 for landscape stuff, but definitely 1600+ for Milky Way and low-light portrait work.
David Graham, I will definitely give that a whirl. I've noticed that my 5D3 can pull detail out of its left hockey skate when I wouldn't think it was there in an overexposed image, much more so that my old 5D2. I just thought it was Canon improving things and I'm sure it is, but maybe I have some extra wiggle room on the right in this particular camera. Thanks for sharing that.
Me being something of a newbie to using a DSLR, would you please outline the steps I'd need to take in order to "Expose Beyond the Right (EBTR)." I'm a little confused as to the process.
I'm using a Nikon D-7100.
Thanks in advance,
--P
http://www.clarkvision.com/articles/exposure_latitude-1/
You can seen that the meter +2 stops still gives good color rendition for all colors in the test chart.
The problem with EBTR is one is relying on the meters, and exposure meters can be fooled in many ways, leading to many ways to saturate the image.
A better way to determine exposure is to monitor the color histogram (not just the luminance histogram). Do a series of exposures and monitor the RGB histogram. When the signal hits the right edge on the histogram, is the raw data actually saturated in that channel? If not how far beyond the right edge can one push it before the raw data are really saturated? If the camera histogram is accurate, going even 1/3 stop over right the edge of the histogram results in saturation in that channel (which again may be recoverable by raw converter reconstruction of the signal).
But in low light photography, as I discussed in my earlier post here, one need not go above ISO 1600 on most cameras. At that ISO, one is digitizing so much detail that even if one underexposes a stop then boosts the signal in post processing there is no loss. Thus there is no need to worry about ETTR or EBTR. To maximize exposure on the subject is not a function of ISO, only exposure time and lens aperture.
The only time I use ISOs higher than 1600 is in low light situations where I need to see the results on the back of the camera just after the image is obtained. For example, I use ISO 6400 for focusing on stars in a night shot, then drop back to 1600 for the real exposure to give more dynamic range.
Roger