[interested_observer]

Several months ago, I was out shooting the Milky Way with the K1 at a new location (which was a bust), but while I was out there (and it was warm), I shot the scene with pixel shift. It was somewhat of a test, I was figuring out the new camera body and I wanted to sort of see how it both worked and what the results looked like.

Well, the location had too much light pollution, just a bad site, along with the location just didn't have any interest - so, I really never did anything with the images. I looked at the out of the camera images, but really didn't pursue any post processing.

So, the Question at hand right now is - Can anyone characterize the quality of noise reduction pixel shift provides in the resulting image? I'm really concerned with the capturing the image aspect here. The reason why I ask, is that in late February, I am planning a blitz of several locations, shooting the Milky Way over some specific landscapes (laying flat across the horizon - only happens once a year). With the new moon and pretty dark sites, it's already a set of harsh conditions (well for me) - but the K1 does soak up a lot of detail at night that I want to use to the best advantage. Since, I'm going to be freezing my butt off anyway, I'm taking around a 7 shot stitched pano anyway, rather than doing a 2 to 4 minute frame (for the foreground landscape shot), I'm considering taking a 2 minute exposure in pixel shift (or 8 minutes for each frame). I figuring that the pixel shift result should be better than just either a 4 or 8 minute long exposure, or several stacked 2 or 4 minute exposures - plus I would have the individual images to boot. The difference is about an additional hour of shooting time (out in the cold). As I stand out there, I just want to be able to think that it's worthwhile. If I'm able to capture the light in the best way possible - I can then always process at my leisure.

Any better approach that I have not considered? In the past I use to ....

• Bracketing +/- 1/3 EV and then stack them (sort of along the line of Exposure Fusion or very light uncooked HDR). This turned out not too bad. I know that bracketing is not a standard noise reduction technique, but it's an easy way to shoot several images with slight variations that stack nicely.
• Stacking several individual images together.
• Using in camera noise reduction - I'm really torn about this. I have not really tried this since the K20D days. Out in the field, I feel that I'm not productive. I'm considering taking a couple of dark frames (perhaps 2 minutes) at the beginning and then at the end of the shooting process, so that I just have them available, if I decide to try something later on. I figure that I can put the lens cover on, put the body/lens in my backpack and let the camera capture the frames while I'm hiking out - since the hardware is cold soaked anyway.

The more I consider the pixel shifting approach, the more I'm drawn to it in terms of noise reduction. I'v been giving it a lot of thought recently. If I'm also able to combine it with a dark frame subtraction (perhaps after the fact in post), that might even have some additional benefits. I'm driving 2.5 hours each way to each location - so this is my make or break attempt, where I really want to move from experimentation to capturing something very worthwhile. Hopefully, the photographer will not be stupid either. One of the locations is reputed to be the 2nd darkest location in the US.

Here is an image from last week (and it wasn't too cold), new moon at 10pm, relatively dark site (60 miles from Phoenix, 5 miles from the nearest small town) - K1, Sigma 18-35 @35mm, f1.8, ISO 800, 2 minutes - single shot, processed in lightroom 5.5. I didn't have the presence of mind to shoot this in Pixel Shift. Obviously, there is noise. Been thinking about perhaps driving back out to shoot a couple of test frames with Pixel Shift to see how it turns out. The post-processing will really determine the actual outcome. There is the Pentax utility (SilkyPix), Lightroom (I believe it now does the pixel shift processing in ACR/LR6), as well as RawTherapee.

• PIXLS.US - RawTherapee and Pentax Pixel Shift - addresses the benefits with a section on less noise.

• Pentax K-3 II Review - Processing Pixel Shifted RAWs | PentaxForums.com Reviews - Touches on noise briefly.

I really have no experience with either SilkyPix or RawTherapee. Any suggestions on which utility that would provide the best results with these types of images? Anyone know if any additional links that address this aspect. There was a super resolution utility (no longer available or supported) that also touched on noise reduction, but there is not a lot of available information.

Where I'm going with this is - bottom line, I want a large (20" x 30") printable image to hang my wall. I should be capturing a resulting image on the order of 15,000 by 7,000 pixels. I've futzed around with shooting astrolandscaping for 4 years now, trying to make due with a cropped sensor (K5IIs) with an excellent lens (the sigma 18-35) - couldn't make it work to my liking. I threw up my hands and decided to go with the K1 in the middle of last year, if I'm serious about this.

Attached Images

 

---

Last edited by interested_observer; 01-22-2018 at 06:06 AM.

[Gimbal]

Pixel shift and the Milky Way, I don't think that will give you anything but head ache. I would stay with normal stacking.

[Bob 256]

One problem with long exposure pixel shift is that stars will have moved between exposures negating any advantage there (possibly creating artifacts) . Foregrounds may be improved since they don't involve motion, but I think I would follow Gimbal's advice and stick to stacking conventional exposures which can offer considerable noise reduction gain.

If you don't want star trails, you might need a guided mount (equatorial) to get the stars right and combine that with a static ground shot. Since you want good details, you need to close your lens to its "sweet spot" and that will require longer exposures which will blur the stars (don't close the lens too far however as that will dim the stars excessively).

[photoptimist]

First pixelshift will significantly improve the landscape part of the shot but it will be useless for the moving stars unless you put the camera on a very good tracking mount.

Overall, pixelshift has two major effects on image noise. First, by quadrupling the number of pixels measured in red and blue, the size of color noise grains will be about half what it is for a regular single-shot image. Second, by averaging 2 measurements of the green channel for every pixel, the luminance noise will be much lower. That is, the noise will be much finer in grain and lower in magnitude.

The reason to do long shutter NR is to control for how the dark frame noise changes as the sensor warms up during use. But maybe it is overkill and your plan to use before and after frames is good enough. However, maybe that one special night of shooting isn't the time to find out if before and after frames are good enough (or not). Before you decide about darkframe subtraction, you might want to try an experiment: turn-off long-shutter NR, cold-soak the camera outside at night, put the lens cap on, take a dark frame, keep the lens cap on, take a pretend panorama series with your chosen ISO, shutter speed, and shot-to-shot movements, then take a second darkframe. Now download all those images and look at the noise patterns in the pretend panorama and whether the before-and-after darkframes are good enough. Maybe the sensor warms up so gradually and so uniformly that the before & after frames can be blended to create a good dark frame for each panorama image. But maybe one part of the sensors warms up faster and creating the correct darkframe is much harder in post.

[interested_observer]

Pixel shift and the Milky Way, I don't think that will give you anything but head ache. I would stay with normal stacking.

One problem with long exposure pixel shift is that stars will have moved between exposures negating any advantage there (possibly creating artifacts) . Foregrounds may be improved since they don't involve motion, but I think I would follow Gimbal's advice and stick to stacking conventional exposures which can offer considerable noise reduction gain.

First pixelshift will significantly improve the landscape part of the shot but it will be useless for the moving stars unless you put the camera on a very good tracking mount.

Overall, pixelshift has two major effects on image noise. First, by quadrupling the number of pixels measured in red and blue, the size of color noise grains will be about half what it is for a regular single-shot image. Second, by averaging 2 measurements of the green channel for every pixel, the luminance noise will be much lower. That is, the noise will be much finer in grain and lower in magnitude.

The reason to do long shutter NR is to control for how the dark frame noise changes as the sensor warms up during use. But maybe it is overkill and your plan to use before and after frames is good enough. However, maybe that one special night of shooting isn't the time to find out if before and after frames are good enough (or not). Before you decide about darkframe subtraction, you might want to try an experiment: turn-off long-shutter NR, cold-soak the camera outside at night, put the lens cap on, take a dark frame, keep the lens cap on, take a pretend panorama series with your chosen ISO, shutter speed, and shot-to-shot movements, then take a second darkframe. Now download all those images and look at the noise patterns in the pretend panorama and whether the before-and-after darkframes are good enough. Maybe the sensor warms up so gradually and so uniformly that the before & after frames can be blended to create a good dark frame for each panorama image. But maybe one part of the sensors warms up faster and creating the correct darkframe is much harder in post.

Folks - I'm ONLY concerned with the landscape foreground segment of the composition, as I'm shooting the sky with the K1's Astrotracing (1 minute) capability. Yes, I sorta skipped over that aspect in the post. As you all know, when using the astrotracing capability, it's going to blur the non-moving items - i.e. the landscape. Since, I need to shoot the stationary foreground again anyway (for compositing), I am looking for the best approach for noise control.

[Rondec]

Pixel shift can help, but only if there isn't much movement in your scene. I do not typically shoot night scenes with pixel shift because with longer exposures, typically you have enough movement in various parts of your scene that it gives little benefit.

[jbinpg]

Yes, pixel shift leads to superior noise reduction, moreso if stacked. The only negative is the increased file size, processing time and storage needs.

[interested_observer]

3 hours of whacking cactus is more than enough for a day....

Pixel shift can help, but only if there isn't much movement in your scene. I do not typically shoot night scenes with pixel shift because with longer exposures, typically you have enough movement in various parts of your scene that it gives little benefit.

The image in the original post, yes - there is the potential of movement. That said, the 4 or 5 times I have been out there, there has been really no breeze. It's an ok dark location, but not the darkest. The absolute darkest location (2nd in the US) is out at the Kofa Wildlife Refuge (the old King of Arizona gold mine up in one of the canyons), and there will be absolutely no movement there (well, if there is, we will all be in a lot of trouble since the entire mountain range is a volcanic plug). In this image just replace the clouds with the Milky Way laying across the entire horizon just resting on the mountain tops.

Yes, pixel shift leads to superior noise reduction, moreso if stacked. The only negative is the increased file size, processing time and storage needs.

That leads to the question at hand - which would be the best

1. 4 - 2 minute pixel shifted images or
2. 4 - 2 minute exposures stacked or
3. 4 ( 4 - 2 minute pixel shifted images) stacked - I don't think that with 7 frames I could hold out in the cold and not enough time before the astro sunrise (which in and of itself would lend an interesting light - sun rise would be in back of the main mountain). This option wouldn't work - I really only have a 2 hour window between MW rise and astro sun rise.

[jbinpg]

Take your four pixel shift exposures. Then you can decide to stack or not later. In my experience, stacking is beneficial, such that a dozen pixel shift shots should result in a noiseless stacked image.

[interested_observer]

Take your four pixel shift exposures. Then you can decide to stack or not later. In my experience, stacking is beneficial, such that a dozen pixel shift shots should result in a noiseless stacked image.

It all comes down to available light, or in a dark location, with no moon, shooting into the source of most of the available starlight (the Milky Way), just capturing whatever other ambient light is available to try to highlight the characteristics in the landscape. Something to bring out some interest other than just having a silhouette.

• It takes about 15 minutes to shoot 2 sets of astro tracked images for the sky (stitched pano of 7 images @ 60").
• It will take 60 minutes to shoot a single set of 7 stitched pano frames of 4 - 2 minute pixel shifted images.

I can shoot the landscape segments....

• before the MW rises, but the source of starlight will be even less, so it will be even darker
• after the MW rises, but the sun/moon will start to rise, back-lighting the landscape
• a combination of one set before and one set after.

The available time window to shoot within this 2 week time span is pretty tight

• The first day - it's a 30 minute window where the MW rises, and then the sun starts to twilight, no moon all night long
• The last day - the moon sets early morning, 15 minutes later the MW rises, available for almost an hour before the sun starts to twilight.

It's also gated on just how much cold I can stand, night after night, even bundled up as an Eskimo.

Trying to somewhat match the various geographical locations with best orientations and time windows - then it's all up to the weather, cloud cover, wind, etc. I figure I'll be weathered out half the nights. Windless nights will have to be with sites with things that can move - i.e., the windmill.

It all comes down to the luck of the draw, even with all the pre-planning, just using the information to help with an intelligent playing it by ear. At least I can roll forward what happens early into the later nights in the time-frame. It's all somewhat of a crap shoot - just trying to stack the odds in my favor as much as possible and then not doing anything stupid.

Out of the 14 days and 6 locations - I'm hoping to be able to shoot all 6 and to get something potentially usable from each one. Then hoping that one of them will be really excellent and printable.

[MJKoski]

I see no problems using pixel shift with good tracking mount. Without one I see problems (multiple exposures of the same stars in different colors in different positions). But, do not use in-camera noise reduction with pixel shift, it totally kills any details gained.

[photoptimist]

It all comes down to available light, or in a dark location, with no moon, shooting into the source of most of the available starlight (the Milky Way), just capturing whatever other ambient light is available to try to highlight the characteristics in the landscape. Something to bring out some interest other than just having a silhouette.

• It takes about 15 minutes to shoot 2 sets of astro tracked images for the sky (stitched pano of 7 images @ 60").
• It will take 60 minutes to shoot a single set of 7 stitched pano frames of 4 - 2 minute pixel shifted images.

I can shoot the landscape segments....

• before the MW rises, but the source of starlight will be even less, so it will be even darker
• after the MW rises, but the sun/moon will start to rise, back-lighting the landscape
• a combination of one set before and one set after.

The available time window to shoot within this 2 week time span is pretty tight

• The first day - it's a 30 minute window where the MW rises, and then the sun starts to twilight, no moon all night long
• The last day - the moon sets early morning, 15 minutes later the MW rises, available for almost an hour before the sun starts to twilight.

It's also gated on just how much cold I can stand, night after night, even bundled up as an Eskimo.

Trying to somewhat match the various geographical locations with best orientations and time windows - then it's all up to the weather, cloud cover, wind, etc. I figure I'll be weathered out half the nights. Windless nights will have to be with sites with things that can move - i.e., the windmill.

It all comes down to the luck of the draw, even with all the pre-planning, just using the information to help with an intelligent playing it by ear. At least I can roll forward what happens early into the later nights in the time-frame. It's all somewhat of a crap shoot - just trying to stack the odds in my favor as much as possible and then not doing anything stupid.

Out of the 14 days and 6 locations - I'm hoping to be able to shoot all 6 and to get something potentially usable from each one. Then hoping that one of them will be really excellent and printable.

Sounds like a good plan.

The one change I'd think about is whether there's a way to capture the landscape with some sort of directional lighting. If you only have the diffuse lighting of the stars for a light source, then the landscape is going to look pretty flat (not unlike what you get on an overcast day). Shooting the landscape portion of the composited shot when the moon (or sun) is off to one side in the sky would create more dramatic lighting of the landscape.

I'd even think about capturing the landscape portion of the shot during the day when you can use very low ISO, the best aperture, short shutter times, and pixelshift. Either during the shot or in post, you can drop the exposure of the landscape until the daytime blue sky is nearly as dark as the night sky to get a more natural "night time" look and to balance the relative darkness of land and sky. The brightness of the blurred landscape of the astrotracked night images can also help you guagethe correct relative brightness of land and sky for a "night".

Best wishes for your photos!

[biz-engineer]

On my K1 copy, pixel shift drifts over long exposures (more than 8 seconds exposure, camera is not able to keep the sensor in the exact same position for each of the 4 pixel shifted frames), the resulting pixel shifted assembly is full of artifacts (I've done a number of test with perfectly static subject such as stones of a castle and sturdy tripod). I wouldn't recommend to use P.S for astro. I'd recommend align and stack multiple exposures of 2 minutes.

[MJKoski]

Even slight changes in light causes artifacts with static subjects. I never had issues with 30 sec PS exposures otherwise.

[interested_observer]

Sounds like a good plan.

The one change I'd think about is whether there's a way to capture the landscape with some sort of directional lighting. If you only have the diffuse lighting of the stars for a light source, then the landscape is going to look pretty flat (not unlike what you get on an overcast day). Shooting the landscape portion of the composited shot when the moon (or sun) is off to one side in the sky would create more dramatic lighting of the landscape.

I'd even think about capturing the landscape portion of the shot during the day when you can use very low ISO, the best aperture, short shutter times, and pixelshift. Either during the shot or in post, you can drop the exposure of the landscape until the daytime blue sky is nearly as dark as the night sky to get a more natural "night time" look and to balance the relative darkness of land and sky. The brightness of the blurred landscape of the astrotracked night images can also help you guagethe correct relative brightness of land and sky for a "night".

Best wishes for your photos!

The knowledge, comments and suggestions that has come out from this thread have been excellent and helps a great deal. The locations I'm planning on using are pretty diverse in general with the commonality of having some type of large lump of rock for the Milky Way to lay across. It only happens once a year, so it's not really something that you can "practice" a lot with. Lighting the entire area is impractical, however lighting a single aspect has been used in a couple of the locations. In a couple of the locations there would be something available to very subtle to light (with very little light - just a couple of lumens). In other locations, there is just a bunch of nothing.

• Kofa - Just a lot of nothing. If I climb a small cinder cone (200' high - very easy in the day) in the dark (something I'm not really wanting to do - if something goes wrong, I'd be toast - hope my wife isn't reading this). If I stay down on the flat (a lot easier, much more prudent) then there are gazillions of teddy bear jumping chola cactus that could be lite from the side - but it would look very artificial. Been considering some aspect similar to this, and been scouring the google earth overheads for something that may work out. But, in that this is the 2nd darkest location in the US, using some light would just not be right. I just might punt here and go with what ever is available. It is what it is.
• Picket Post (windmill image in the OP) - The windmill is the obvious opportunity here. It would be probably best from underneath shooting up due to its height, with a couple of lumens. There is some light from a small town 5 miles away and other light from an open pit mine 30 miles away. I've found that even the light from Phoenix, 60 miles away - provides a fair amount illumination on the mountain face.
• Lost Dutchman - There are Saguaro Cactus that would work, and also a gate in the cattle fence that could be used. The cactus would work better. There is some ambient light from a small town a couple of miles away that reflects off the the rock faces of the mountain pretty nicely.
• First Water - There is a dry wash that could be very subtle from one side that could be pretty dramatic. On the other hand, there is sufficient ambient light from a small town a couple of miles away that lights up the rock faces pretty nicely. I did a throw away shot here a couple of years ago, that turned out way better than my expectations.
• Four Peaks - Again, just a lot of nothing for miles and miles and miles. This may work out, and may not since the landscape element is really waaaay off in the distance. Working on a couple of different locations that offer different aspects on this one.
• The Goldies - 500 foot rock faces about a mile long across a river a half mile away. Still working out the angles, and the specific location to shoot from to provide the desired angles and views. This is the closest location to Phoenix that has a fair amount of pollution from the light dome, so that would provide more reflected light off the rock faces that depending on conditions will either work with or against the desired outcome.
• midgley bridge sedona - Lighting up the bridge in a very subtle way can work - along with the car lights crossing the bridge every now and then. There is a small amount of light pollution from Sedona (a designated dark town), that would provide ambient lighting off the rock faces a couple of miles away.

So, all in all, I have taken this into account to varying degrees. Some of it is just what ever ambient light pollution is just there. The tough part is that there is just enough to help a bit but also brings out the noise in the image enough to be annoying. These locations for the most part are sufficiently dark (even considering whatever illumination) that it takes some very long exposures to make things work.

On my K1 copy, pixel shift drifts over long exposures (more than 8 seconds exposure, camera is not able to keep the sensor in the exact same position for each of the 4 pixel shifted frames), the resulting pixel shifted assembly is full of artifacts (I've done a number of test with perfectly static subject such as stones of a castle and sturdy tripod). I wouldn't recommend to use P.S for astro. I'd recommend align and stack multiple exposures of 2 minutes.

The Astrotracing really works best for the sky part of the image. It's the landscape part that I want to collect enough light to bring out enough of the landscape to be interesting and reduce the noise enough to maintain that interest.

______________________

After getting the K1 I was going to order a 15-30/2.8, but my wife somewhat objected - "you just got a new body and NOW you want a new lens too". After shooting some test shots the other week - she asked how they turned out. I said I got 4 images - and she replied, you were gone for 6 hours and all you got were 4 (99 images stitched into 4 results), and she said - why don't you just get a panorama lens. The new 15-30 just arrived yesterday (I broke the piggy bank - I was saving just for this for quite a while).

I have been using the Sigma 18-35/f1.8 for the last few years. I used the lens for the test shots at f1.8 with the K1. The problem is even at 35mm where you get full coverage of the full size sensor, when you stitch everything together, the center of each individual image is reasonably brighter than the edges (a very subtle form of vignetting - even with all traces of vignetting removed by LR). So you get somewhat of a patch work quilt. Going to the 15-30 fixes that, but I loose a 1.3 stops of light. Just another aspect to work into the equation. It is what it is, I will just capture what ever light there is - hoping not to do anything stupid - and just roll with it.