1. Reminder: Please use our affiliate links for holiday shopping!

OMD Long Exposure Noise Solution

Discussion in 'Olympus Cameras' started by tradesmith45, Jan 10, 2016.

  1. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    In my last post, I explored some ways of reducing long exposure noise (LEN) but made only modest progress: Long Exposure Noise - Parameter That Improve the E-M5

    I started this because I enjoy making night landscapes & those sometimes require exposures >1 min. But OMD cameras produce significant noise at ISO=>1600. Here's a capture I did at the Grand Canyon w/ a histogram of just the red rock walls to illustrate what min. exposure is needed for star light only work.
    23588154354_7502b910c9_b. RedRockhistogram.jpg by tradesmith45, on Flickr
    So ISO 3200, f2 & 2 min. is what's needed but for my Kowa f2.8 that would be 4 min.

    There is substantial noise in this image that will limit print size to something rather small. Certainly the rumored f1.2 WA lenses will help this problem if they are sharp enough. But even those lenses will leave OMD cameras at a distinct disadvantage to Fuji & a few Canon & Nikon DSLRs. See the stunning photo Roger Clark has on his home page made w/ a 6D for example: Clarkvision.com home

    I wanted to try several options for reducing this noise. To test, I created a simple "scene" in my office that I could light w/ an adjustable LED head lamp placed in black nylon cloth. This gave me dim & somewhat reproducible light - except when stray light came past the window curtain from say a passing car. (All the following images are 100% crops w/ typical LR adjustments: modest increases in Noise Reduction, Contrast, Clarity, Vibrance & Sharpening - the kinds of modest adjustments I'd be likely to make to an image I'd like to print.)

    First I tried stacking 9 images in PS & then dark frame subtraction in post processing. As you can see, neither worked very well.
    23848523089_40ea98569c_b. 9xStack 3200.jpg by tradesmith45, on Flickr
    23920786050_4d2b08ea68_b. DFS Post 400.jpg by tradesmith45, on Flickr

    Then I tried using very high ISO to shorten the exposure time needed:
    24190274676_5a54970651_z. ISO5000.jpg by tradesmith45, on Flickr

    This too failed. And post processing noise reduction software worsens the visibility of LEN by turing color pixels to white.
    24129016172_9b42217dd4_b. LR vs OV3.jpg by tradesmith45, on Flickr

    As mentioned in my first post, the E-M5mII has more long exposure noise than the mkI:
    23848541889_0223149ef3_z.

    And the E-M5mi get really bad if you have to use 4 min. to accommodate an f2.8 lens.
    EM5_Mk1_vs_2_ISO_3200.jpg by tradesmith45, on Flickr
    24216392685_bf4f972057_z. M5II_NR0ff3200.jpg by tradesmith45, on Flickr

    But I had found that turning on in camera Noise Reduction for the Mark I would significantly reduce long exposure noise:
    23588159344_eee383de09_b. EM5NrOn3200.jpg by tradesmith45, on Flickr

    Having just bought a markII, I did extensive testing to see what is possible with this camera. Generally, turning on in camera Noise Reduction significantly reduces LEN but leaves some fine black grain in the picture- more for higher ISO &/or longer exposure durations. Haven't printed anything yet but the grain at ISO 3200 for 2 min. looks acceptable. Not sure ISO 3200 & 4 min. is acceptable however. Here's result for equal exposures @ ISO 3200 @ 4 min & 1600 for 8. Not sure you'll be able to see this but ISO 1600 is better but to use it, the camera will be tied up for 16 min-ugh.
    23848555029_765f926b7b_z. NRon3200vs1600.jpg by tradesmith45, on Flickr

    So I also tried ISO1600 & 8 min. but w/ NR off.
    23589561073_12f43f755f_z. NR3200vsOff1600.jpg by tradesmith45, on Flickr
    Not even close.

    But ISO1600 @ 4 min.w/ camera NR off isn't too bad but longer exposures even at ISO 800 have more noise.
    24133810931_16862a8575_z. NRoff1600800.jpg by tradesmith45, on Flickr
    So reducing ISO by a stop (& doubling the exposure time) does NOT substitute for using in camera Noise Reduction, its still needed.

    As mentioned, this problem applies to star light only night shooting. So can the E-M5mII do well if there is supplemental lighting? Yes, here's results for exposures needed is the scene is lit by a quarter moon (50%) & in camera NR turned off. Virtually no LEN visible:
    24108325962_e5260db76a_z. ShortSSEM5II.jpg by tradesmith45, on Flickr

    One last thing regarding pixel mapping. All the OMDs have this function. Its basically a test so the camera can ID the stuck/hot/bad pixels on the sensor & create a map of their locations. This is different from LEN however. The camera then interpolates a signal for these bad pixels by sampling from surrounding good pixels. Roger Clark says for at least Canon, interpolation done during RAW development (probably by placing the bad pixel map into the EXIF data) so in camera noise reduction isn't needed. Sure wish Oly did that. From these tests, I've concluded that the only way the bad pixels map gets used in an OMD is is in Camera NR is turned on & this function is much more than just simple DFS.

    Where this leaves us is needing big lens apertures & really long exposure times to keep exposures below ISO 3200 for 2 min. Fuji is so much better at this! I'm still thinking about the MZ 8mm f1.8 too.
     
    • Like Like x 6
    • Informative Informative x 4
    • Appreciate Appreciate x 1
  2. atarijedi

    atarijedi Mu-43 Regular

    69
    Dec 13, 2015
    Have you tried just stacking a lot of low iso/short exposure images, like ISO200 and 10s shutter?

    Also, just as something to test out, what is high-res mode like for astrophotography? I wonder if it would even work. If you could get a clear exposure, maybe you could interpolate them down, and stack them (or vice versa). That would get rid of noise.

    Although neither of these would work for Aurora, since's it sorta moves around like a flag in the wind. Here's hoping the EM1mk2 has better low light noise performance!
     
  3. wjiang

    wjiang Mu-43 Hall of Famer

    Short exposures are no good as you'll need to push them way too much. We are talking really dark here - ISO200 10s would be almost pitch black, you'll need much higher ISO than 200. Hi res mode might work if your lens is fast enough - the longest shutter length is 8s and maximum ISO is 1600, but noise is likely to be even more of a problem due to electronic shutter.
     
    • Agree Agree x 1
  4. atarijedi

    atarijedi Mu-43 Regular

    69
    Dec 13, 2015
    I think the better option is to keep a low ISO, since it will simply amplify any noise present, but extend the exposure time. Keep the ISO at 200 or 400, and go for say, 80s or 40s exposures respectively, or however long is needed. Obviously you'll need a star tracker at this point, to prevent trails.

    You could even use LiveBulb or LiveTime to determine the optimal exposure length, although the LCD can't really be trusted, as it supposedly shows more noise than what is present in the image. I also remember reading you should capture a dark frame every 3 to 4 exposures.

    I'd also like to see an E-M1 comparison (the camera I have), since it hasn't been done yet.
     
  5. wjiang

    wjiang Mu-43 Hall of Famer

    By all accounts (there's plenty of examples online) the E-M1 is horrible for long exposure noise. I did a quick comparison between my E-M1 and GM5, and for ISO200 the E-M1 became noticeably worse than the GM5 for exposures longer than about 1 minute. I get by stacking something like 8-16 30s shots at ISO1600 with the 7.5FE at f/3.5, but in camera noise reduction (DFS) must be on. With it off, I find it unusable for long exposures.
     
    • Agree Agree x 2
  6. Klorenzo

    Klorenzo Mu-43 All-Pro

    Mar 10, 2014
    Lorenzo
    Just because you did not mention this in your post. For Olympus:
    - noise reduction = dark frame subtraction, applies to RAW too
    - noise filter: the common denoise on jpeg only (default setting is a little strong, low/off could be better).
     
  7. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    You've got the distinction between the two right but I no longer think "noise reduction" is simply DFS. Here's why.

    First even after doing Pixel Mapping several times on my E-M5, all my long exposures always contained many stuck pixels in the same locations + lots of other thermal noise. I concluded the Pixel Map is not applied to long exposure RAWs when Noise Reduction is turned off.

    Second both the stacked images & the DFS done in post all show lots of large black grains of various shapes. Subtracting the noise contained in the dark frame or merging many frames simply creates blanks (black) where noise had existed.

    Something completely different happens when Noise Reduction is turned on. The images are amazingly clean. My conjecture is that turning Noise Reduction on causes the camera to use the dark frame not for subtraction but rather to impute a signal for the noisey pixels from adjoining pixels free of thermal noise. Doesn't work perfectly every time because there are sometimes still hot pixels. And imputing a signal from surrounding pixels brings both the signal & the shot noise from the adjoining pixels.

    What ever is actually being done under the hood can not be duplicated in post by simple subtraction or stacking.

     
    Last edited: Jan 11, 2016
    • Informative Informative x 1
  8. TonyVentourisPhotography

    TonyVentourisPhotography Mu-43 Regular

    68
    Nov 1, 2015
    Being in astrophotography myself...I keep seeing insanely good results (in terms of oly cameras, so relatively speaking) from the e-pl5. Apparently this is even better than the e-m5 mk I. There is a company even offering an astro-ready version of that.
     
  9. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    Interesting. Do you have an E-PL5? If so, it would be helpful if you could do a long exposure test so we can compare.

    As far as I can tell, this model has 5-axis IBIS like others & has the same processor. Don't know about sensor differences. So its hard to see any physical reason the E-PL5 would be any better than the E-M5.

    I'm not surprised someone (who?) offers astro converted E-PL5 given its low cost.

     
  10. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    There are significant practical limits to how low ISO you can use. The exposure time needed for star light scenes w/ f2 & f2.8 an ISO 400 would be 16 & 32 min. (but 30 min. is max). Add in camera Noise Reduction & it would take 32 to 60 min. for a single exposure. I did test ISO 400 w/ NR On & the results are certainly clean. So if you are willing to wait that long, it does work.

    I also test ISO 800 both w/ & w/o Noise Reduction. Even @ ISO 800, 8 min. long exposure noise was serious unless Noise Reduction is turned on. 16 min. would be even worse. I didn't test it, but I'm confident that ISO 400, 30 min. exposures w/ NR off will have lots of noise.

    So turning on NR is required for all minutes long exposures & that means ISO 1600 is probably the best option but will still require 8 & 16 min. for f2 & f2.8 lenses.

     
  11. Lcrunyon

    Lcrunyon Mu-43 Top Veteran

    758
    Jun 4, 2014
    Maryland
    Loren
    Thanks for all of the exhaustive testing. It's very informative, and must have taken a lot of time. Appreciated.

    I'd love to try astrophotography and starry night landscapes some day, perhaps with a little subtle light painting to reduce exposure time. The exposure times you are discussing will obviously produce long star trails, unless you have one of those mounts that moves with the earth's orbit. If you wanted to do tack-sharp images of the Milky Way, my understanding is that the limit to how long you can expose is pretty short (best done with fast wide angle lenses). Is E-M1 or 5 (I or II) noise as bad in those situations? Are there particular ways it should be handled, or just Noise Reduction On and whatever ISO is needed to get the maximum exposure time? I've seen some nice images like these from folks so I assume they are within the OM-D's technical constraints.

    Also, your post makes me think I should look into pixel mapping. I haven't yet noticed any bad pixels, but I might be surprised.
     
  12. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    Loren, lets separate your question into 2 parts- sky & landscape. Generally I feel that m43 can do decent start landscapes but you have to work harder (use multi image techniques, a star tracker or have supplemental light) to get there.

    Here's a single frame of the MW w/ mild NR by DXO Prime & in camera NR turned off: E-M5, MZ 12mm @ f2, ISO 3200, 25 sec.
    24366010411_b8ee2aa7c2_b. _8120018DXO.jpg by tradesmith45, on Flickr
    As you can see, the red gas cloud in the center is certainly visible along w/ other colored objects (& some fringing).

    This image is plenty acceptable for the sky in a landscape but I've gotten even better results by stacking 5-8 images. This lens is remarkably good for such a short FL. There may be only 1 or 2 lenses that are f2 & 24mm Eq-FL that are as good. If the rumored Oly f1.2 is as good, it will be a huge advance for starry landscapes even if it has to be stopped down to f1.4. For stacking I'm finding Starry Landscape Stacker works great but its Mac only. Deep Sky Stacker is favorite in the Windows world.

    The landscape part of the image is where the really long exposures w/ in camera NR turned on are required and that's what motivated me to do these tests. But again, I'm hoping the new f1.2 primes will be a help.

    As for cameras, as Allan points out the best astro images are from cooled sensors. But as images from many folks have shown starry night landscapes are certainly possible w/ the best of the m43 cameras. So far as I've seen, the E-M5 is the best performer but at least one person said the E-PL7 is better. And DPR found the E-M10 Mk2 has less noise than any of the other Olys.

    But I'm still keeping my eye on Fuji - great lenses & very low noise for short or long exposures. For this kind of work, there is no contest, Fuji beats Oly. But starry landscapes are not the only type of photos I take.


     
    • Appreciate Appreciate x 2
  13. ExCoderAtWork

    ExCoderAtWork Mu-43 Rookie

    15
    Jan 20, 2014
    Here is a post I did last July comparing the E-PL5 to the E-M1 for long exposure. The difference was unbelievable. The E-PL5 on has 2-axis IBIS. Not sure if that runs any cooler than the 5-axis in the E-M5.

    Long Exposure Comparison
     
    • Like Like x 1
  14. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    Nice work, seems like several of us are dealing w/ this problem.

    Would love to compare the E-PL5 to the E-M5 for long exposure noise to see if the simpler IBIS is better. If you wouldn't mind doing a couple captures w/ your E-PL5 & putting the RAWs in DropBox, I'll do the comparisons & post them here. All you need to do is put on the body cap, have the camera at room temp (70), make exposures @ ISO 3200 for 2 & 4 min. & Noise Reduction turned off. Hope you can help.

    From looking a Brendan Davey's tests, I suspect there are multiple factors that cause increased sensor heating during long exposures: number of pixels, number of AF sensors, sensor shift IS & how much heat sinking is used. There are lots of cameras using various Sony sensors & their long exposure noise performance ranges from terrific to terrible. Olys are on the terrible end. There are large differences among the different Sony A7 models to. All the best performers (6D, D750, Df, Fuji X, A7S) do not use sensor shift IS (which almost certainly limits heat sink size) and have fewer total pixels.

     
    Last edited: Jan 19, 2016
  15. ExCoderAtWork

    ExCoderAtWork Mu-43 Rookie

    15
    Jan 20, 2014
    Sure thing. I will try to do them tonight and send you a link to the raw files.
     
    • Appreciate Appreciate x 1
  16. Lcrunyon

    Lcrunyon Mu-43 Top Veteran

    758
    Jun 4, 2014
    Maryland
    Loren
    If you are doing such long exposures, wouldn't you have a tripod and IBIS turned off? Does IBIS still put off heat when not in use?
     
  17. wjiang

    wjiang Mu-43 Hall of Famer

    It's not so much the IBIS being on. It's the fact that the whole sensor block is being held up by voice coil magnets and not attached to significant chassis heat sinking. Even with IBIS off it is just floating there (held static by the voice coil magnets), attached to the rest of the body only by flex cable.
     
    • Informative Informative x 3
    • Agree Agree x 2
  18. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    Thanks to ExCoderAtWork, here's a comparison of the E-M5 & E-PL5. As you can see the E-M5 has less LEN. Was wondering if the simpler IBIS made a difference. Doesn't seem to help.

    Sure wish someone w/ a GH4 would join in this testing.

    P1010002.jpg by tradesmith45, on Flickr

    Been looking at Spencer's Camera & Photo conversion service to get a full spectrum conversion. They offer a "heat reduction system" for an additional $125. They say it helps but not as much on the IBIS equipped cameras. Anyone out there tried this?
     
  19. tradesmith45

    tradesmith45 Mu-43 All-Pro

    Dec 13, 2012
    Oregon
    Hi LEN Followers. A friend who just got a GH4 did the usual test for me. Sadly the lack of IBIS did not make the GH4 a solution for long exposure noise with m43. Just for comparison also tested the E-M1. Still looking like IBIS is part of the problem. But the E-M5 is still the m43 noise champ.

    Tomorrow, I'm sending my E-M5 off to Spencer's Camera for a full spectrum mod & for his heat reduction system. Don't expect it to help much but will report what I find.

    24777782436_dde8cbdce7_o. M5 GH4 M1 2min.jpg by tradesmith45, on Flickr

    24686192742_f6ea6e0a72_o. M5 GH4 M1 4min.jpg by tradesmith45, on Flickr
     
    • Like Like x 1
  20. wjiang

    wjiang Mu-43 Hall of Famer

    Wow the E-M5 is crazy good by comparison... wish I'd know that before I replaced it with an E-M1!