[MarcusCalidus]

Just wondering if it makes a difference if you do a real long exposure with ND filters or if the Average Interval Composite mode (K-3 / K-1).

I once read somewhere that the noise would increase the longer the exposure takes. So I guess the Average Interval Composite mode would be the go to choice?
Then however since the minimum interval time is 2 seconds you'd loose the trail of slow moving objects.

What is your insight about that matter? What technique do you prefer?

Just curious if anyone did trial testing on that matter before.

[Cerebum]

I was just reading about this in the KP handbook. Great minds it will be interesting to see the take on this. Meanwhile I am off to google it

[biz-engineer]

Long exposure vs Average Interval Composite

Very interesting post.

I once read somewhere that the noise would increase the longer the exposure takes. So I guess the Average Interval Composite mode would be the go to choice?Then however since the minimum interval time is 2 seconds you'd loose the trail of slow moving objects.

Hum yes. In conposite stacking there's a trade between exposure time and the time lapse between two frames where the continuity of subject motion is lost such that composite doesn't render as smooth as a single longer exposure. For example with moving water, stacking approximates a smooth exposure only with a large number of frames stacked ( x 10 exposures or more). For slow moving clouds , the time lapse between exposures (set to "minimum" in the K1) is negligible relative to cloud motion.

A few things to consider:
- I don't think that stacking improves dark current noise of long exposures because dark current noise is proportional to exposure time but it is mostly not random so it is not eliminated by the stacking. Long exposure noise reduced (LENR) is not recommended in a composite capture because of half time require to capture and remove dark frames. If LENR should be performed on a stack, it is preferable to take a black exposure with the lens cap on, and perform LENR on each frame before stacking at post processing (adding frame off-camera).
- composite stacking bring the benefit of lowering read noise, so stacking 2 shots at ISO100 is more like getting better than the image quality of an ISO50 sensor. Stacking improves read noise so much that it can exceed the max S/N of 8bit JPEG which produces banding in skies or any image area with low change in tonalities.
- frame alignment is not performed with in camera composite exposure, any camera sensor micro-displacement induced by mirror and shutter up-down slightly reduce the resolution of the stack
- if the shutter speed is too quick without ND filter, the number of frames required to emulate a long exposure may be an overkill, better use a weak ND in that case.

So typically, stacking 4 to 8 images in live view mode, at ISO100, produces stunning image quality. More than 8 frames stacked create banding in JPEG (unless noise is added again when exporting the RAW to JPEG).

I use composte so that I carry only two ND filters for long exposures. I use a weak ND filter and if the exposure isn't long enough I stack frames. If the stacked frame total exposure time is not enough I use a 10 stops ND and if not enough I use multiexposure again. I hope you find my comment helpful.

[MarcusCalidus]

I hope you find my comment helpful.

Thank you very much for this professional reply. This is indeed very helpful. I did not even think about the problem of the missing alignment of the shots. I guess with a wide angle and fine landscape details this really could pose a problem. I really haven't yet started to try long exposures. But I guess I will have to invest in ND filters then.(some weak at least)

---

Last edited by MarcusCalidus; 10-02-2019 at 03:54 AM. Reason: typo

[biz-engineer]

I guess with a wide angle and fine landscape details this really could pose a problem. I really haven't yet started to try long exposures.

You could try a composite stack (using a remote trigger so that to avoid touching the camera between frames) and compare it to a single exposure, so that you can make your own mind, then decide what kind of ND filter(s) would work for you. For me, there is a small advantage of stacking a number of frames between 4 and 8 frames, it is that I can pull shadow on a raw composite exposure and still have no noise after I pull shadows. Stacking 4 frames improves noise as much as pixel shift, except that no special software is needed for the stacking method, and the stacking doesn't produce any artifact with moving parts in the image.

[MarcusCalidus]

You could try a composite stack (using a remote trigger so that to avoid touching the camera between frames)

Why would you have to touch the camera? With the K-3 there is an automatic mode that takes the shots. The only thing I would have to worry about then is the mirror slap. Then again the K-3 "says" that shake reduction is active even if used on a tripod.
I guess I will have to do some field testing. Alas -- no waterfalls near me

[biz-engineer]

Why would you have to touch the camera? With the K-3 there is an automatic mode that takes the shots.

That's a good point. Yes, you are right, a remote is not mandatory to have the camera take a series of shots on its own, using camera's own interval mode.
One thing I noticed on my K1, is that the interval mode doesn't work with live view electronic shutter mode, on my K1 interval mode revert to DSLR mode (electronic shutter disabled) after the interval is started, while the composite exposure without interval mode keep the live mode running for all exposures of the stack. So I tend to get sharper composite stacks using an IR remote trigger (a 10 Euros JJC device) compared to stacking the same number of images with the interval mode of the K1. Maybe on the K3 it is different, I can't tell because I don't have a K3 anymore.

[MarcusCalidus]

keep the live mode running for all exposures of the stack. So I tend to get sharper composite stacks using an IR remote trigger

that's interesting. I guess that, since the mirror is already locked for the live view, there is no slap shaking introduced. That's a great point. Never thought of that. I will have to try that out. I have IR remote, so that should not be much of a problem.

[CarlJF]

FWIW, in practice I have found long exposure way much easier to do than composite stacking, for all the reasons mentioned above. Although composite stacking works great on paper, it isn’t easy to do correctly and achieve good results. There’s plenty of variables needing to be managed, meaning more chances that something could go wrong.

Although it’s fun to try composite stacking, don’t expect it to replace ND filters which should remain your first choice for long exposure.

[troenaas]

Just wondering if it makes a difference if you do a real long exposure with ND filters or if the Average Interval Composite mode (K-3 / K-1).

Actually, I use both methods. If the lens takes filters, I always use a ND filter. If the lens doesn't take filters, I use composite mode.

[jgnfld]

Just wondering if it makes a difference if you do a real long exposure with ND filters or if the Average Interval Composite mode (K-3 / K-1).

I once read somewhere that the noise would increase the longer the exposure takes. So I guess the Average Interval Composite mode would be the go to choice?
Then however since the minimum interval time is 2 seconds you'd loose the trail of slow moving objects.

What is your insight about that matter? What technique do you prefer?

Just curious if anyone did trial testing on that matter before.

Haven't fiddled with this for a bit and this isn't my best example (best I could find in my albums here), but doing BOTH (ND filter + stacking) in the SAME shot can lead to a nice horsetail hair effect.

Whole pic...


100% crop to show details

[MarcusCalidus]

BOTH (ND filter + stacking) in the SAME shot can lead to a nice horsetail hair effect.

now that looks interesing. Like a bunch of glass fiber
well done

[DeadJohn]

Digital composite works well, but there are limitations. I find that conposite tends to work best in shady conditions that already allow a slightly long shutter speed, so there's already some blur in each frame.

Water flowing at different rates in different parts of the frame can give poor results. For example, a leaf floating in a stagnant part of a stream, while water elsewhere is moving faster, can look odd.

For architecture, composite mode does a questionable job blurring people out of frame.

[MossyRocks]

I've played around with this some and have gone back and forth on getting a set of big dark ND filters or not. I don't really do enough ultra long daylight shots to justify throwing gear at the problem so will frequently do a composite image. What I've found works best is not using the interval composite mode in the camera but shooting in burst mode and then stacking with software outside the camera. Generally my process for doing this is:
1. have the camera on a very sturdy tripod
2. Get as long of an exposure as you can without making a mess of things. I really want to be at 1/10s or longer
3. Put the camera in high speed burst mode, and full manual shooting and use the settings discovered in step 2 that give the longest result without clipping. I leave shake reduction on to deal with mirror slap.
4. Use a release cable and fire away until the camera starts slowing down. Depending on the shutter speed I was at and how smooth I want things to be I may shoot several bursts. I really want at least 30 images at a minimum but like to have 100 or more if I am stuck at a high shutter speed.
5. I usually develop these images in rawtherapee as I like the batch edit mode there better than in lightroom. I upscale to 2x regular size (helps with alignment) and batch export to 16 bit tiff. Do not do noise removal, distortion correction or CA correction at this point if you want to do dark frame subtraction described below.
6. I will align the tiffs using the align_image_stack program from hugin. I usually do a course x/y alignment, fine x/y alignment, a fine rotation alignment, and a final fine x/y alignment. I find I get better alignment this way than photoshop does but it isn't as easy to use
7. I will then load the images into either photoshop or GIMP as a stack and do an average and flatten.
8. At this point you will still have a 2x dimensional resolution image and may want to sharpen it and have a super resolution image that likely has a true 16 bits per channel of data or instead if you don't need that many pixels you can always just down sample to the original image scale.
9. Once you have the flattened stack you can now begin your regular adjustments as you are working with a very high quality clean image that will give you a lot more room for processing than you are use to.

On the subject on dark current noise you can always do what the astro shooters do with dark frames. It is a better version of the slow shutter noise removal. Basically you will need to shoot a pile of images with the lens cap on (dark frames) at the same shutter speed, ISO, and burst speed that you shot the original frames with actual images on them (light frames) right after step 4 above as you want your dark frames to be shot at similar heat and environmental conditions to your light frames. You then develop these using all the same settings you did for the light frames and stack them. The stacking is easy as you don't need to align them just a simple averaging works and now you have a master dark frame to use. Because this dark frame is the average of a big stack the truly random noise should have been pushed down leaving mostly just the dark current noise. If you want to remove this dark current noise you can subtract your master dark frame from each of the light frames between step 5 and 6. A tool like imagemagick makes this easy to do.

With daylight scenes dark current, signal amp noise, systematic AtoD conversion noise, and read noise aren't really big concerns as the signal to noise ratio is pretty high for the image. In astrophotography you have a lot of shot noise (the truly random non systematic noise) with very dark images generally shot at high ISOs so all the other sources of noise actually are a big deal and need to be dealt with. With daylight shots the only one that might be an issue would be the dark current but that may only be slightly noticeable after a really long exposure or on a really old inefficient sensor, which is why when I have done these types simulated ultra-long exposure shots I don't bother doing dark frame subtraction.

I like to tinker and before I got a DSLR played around a lot with trying to maximize the image quality and what could be done with a cellphone camera because I could.

[MJKoski]

Oh yea, average works.

128x 3sec frame average to retain highlights AND allowing about 7 stop boost for dark areas:


60x30sec averaging


Just do not go too far or you will get this:


It is a failed 600x average due to internal buffer overflow and/or calculation precision limit. I found out that 256x is a practical limit for 0EV exposed images containing lots of mid-tones. These were taken with K-1 and K-1 mk2.

Here is about 128x 10sec average with Pentax KP:


Here is a rare example were 512x average worked (technically):


In the image, there are chunks of ice going around in circles due to water movements.