? about sensor processing/encoding/filtering

JC

I'm trying to figure out why there are gaps in multiple shot star-trails.

I know the conventional wisdom posted all around the internet, is that it is due to the gap in time between shutter exposures.

That's not true, you can test that by doing shots of the same period and frequency with different focal length lenses, the distance on the sensor that the stars move is proportional to the focal length, but the gaps don't scale the same way.

Also, the gaps are linear features, and don't align perpendicular to the star motion (see my image below).

The gaps must be due to electronic processing. Either because the shutter closes before an electronic bucket is filled and so less light is integrated at the end of each exposure, but these electronic buckets have a scan line behavior, or it is some artifact of the de-mosaicing of the bayer pixel array-( would a foveon sensor show these gaps?), or is due to some other sensor processing that I'm not thinking of.

I can't seem to work it out.

Anyone here sufficiently versed in the arcane aspects of sensor/computer technology?

This is a crop of one of my images, from the eastern edge of the frame, so the stars would have been moving up and to the right in this context.
The gaps are minimized by stacking subsequent images in screen mode, but it doesn't eliminate them entirely (this is stacked in screen mode).

I've tried turning off the ISO noise reduction in camera (not the long exposure dark frame NR), but that doesn't change it.

NeilL

Facing south?

Do the "gaps" count the same as the separate exposures?

The "twinkling" of the stars is itself like opening and closing a shutter, so when the stars are "off', particularly a number of them "in phase", there would be a gap in the image. So many stars, so many twinklings, might add up to an ad hoc statistical "rhythm". Try going beyond the atmosphere!

This is a short long exposure of a stationary neon sign while I slalomed the camera, analogous to your stars situation. Why the "gaps"?



Neil

JC

NeilL wrote: »

Facing south?

Do the "gaps" count the same as the separate exposures?

The "twinkling" of the stars is itself like opening and closing a shutter, so when the stars are "off', particularly a number of them "in phase", there would be a gap in the image. So many stars, so many twinklings, might add up to an ad hoc statistical "rhythm". Try going beyond the atmosphere!

This is a short long exposure of a stationary neon sign while I slalomed the camera, analogous to your s

Neil

South of east.

That's about 20 or so frames. Gaps occur between frames.

ziggy53

Do you have the "Long Exposure Noise Reduction" turned on?

Is the effect exaggerated at the edges of the frame?

Where is the polar center in the photograph?

NeilL

Don't follow you here. I estimate more than 20 gaps in that image.

The gaps, if I am looking at the right thing, are regular and fast, and as I thought you originally argued, could not be related to frames. So, yes, could be related to sampling rate of sensor (but I would expect that to be much faster than would account for what seen here), or "beats" due for example to the statistical average of star twinkles.

Neil

kolibri wrote: »

South of east.

That's about 20 or so frames. Gaps occur between frames.

JC

ziggy53 wrote: »

Do you have the "Long Exposure Noise Reduction" turned on?

Is the effect exaggerated at the edges of the frame?

Where is the polar center in the photograph?

Hmmm, I'm explaining this badly.
No, I don't have LENR turned on, these are 60 second exposures with 1 second between exposures.
Yes, the gaps are occurring because I am taking multiple images of a moving light, but the explanation is not that it is because of the one second between frames, or at least, that is not a sufficient explanation.

Maybe an equation will help?
F= focal length
T=time between exposures
A=angle away from the equator
G= size of the gap
C= unknown constant

G=T*F*(90/A) +C

-At longer focal lengths, the gap should get larger if it's only a function of time between frames
-If the gap was only due to the shutter being closed, it should get larger the longer the time between frames
-If the gap was only due to the time between frames, the size of the gap should get larger as you shoot away from north.

All these do influence the size of the gap- but, there is an additional variable "C" creating this effect, and that is what I want to pin down.

Here are the side crops again- on them I've drawn an arrow to show the direction of star travel. If the gap was due only to the lost exposure while the shutter was closed, then the gap should be perpendicular to the star motion. It's not.

a) from south of east, the star is rising into the night sky -there are about 17 gaps, so, this was about 18 shots. It's too regular to be 'twinkling'. Besides, a 60 second exposure should integrate all potential twinkles into the average star temperature color.

b) north of west, the star is setting towards the horizon

c) full image, crops are from the right and left side (this was just a fun image while out working- not one of my better images)








What it actually looks like to me is a scan line edge on a georeferenced satellite image caused by the satellite progressing along it's track while it is scanning.

NeilL

Where does your equation come from? A wildcard constant doesn't seem to me to be legit. Its value should be inferred from the requirements of the equation. What is that?

Yes, it's plausibly some kind of "scan" artifact, or sensor sampling rate, tho as I said I would have expected that to be much faster than the visible evidence.

The orientation of the gap lines suggests to me something analogous to a resultant vector. If so, the effect is being created by movement (fluxion-calculus), which doesn't seem to be in your equation, unless it's the constant. Rotation of the sky is a constant, and so is the relationship between that and the periodicity of twinkling averages (if that exists), or perhaps sensor samplings.

In my pic above, the movement of the camera together with the relationship of that to the cycling of the neons, seems to be the origin of the bands in the light. It is not smooth motion blur, which you might expect from sunlight. There are "beats", and beats are the regular cancelling/augmenting of data by data, a "resultant vector"-type value, as I would expect. Same kind of thing as the backwards rotating wagon wheels in the movies where there are clear beats (and possible orientation off the perpendicular to the axis of rotation).

Neil

kolibri wrote: »

Hmmm, I'm explaining this badly.
No, I don't have LENR turned on, these are 60 second exposures with 1 second between exposures.
Yes, the gaps are occurring because I am taking multiple images of a moving light, but the explanation is not that it is because of the one second between frames, or at least, that is not a sufficient explanation.

Maybe an equation will help?
F= focal length
T=time between exposures
A=angle away from the equator
G= size of the gap
C= unknown constant

G=T*F*(90/A) +C

-At longer focal lengths, the gap should get larger if it's only a function of time between frames
-If the gap was only due to the shutter being closed, it should get larger the longer the time between frames
-If the gap was only due to the time between frames, the size of the gap should get larger as you shoot away from north.

All these do influence the size of the gap- but, there is an additional variable "C" creating this effect, and that is what I want to pin down.

Here are the side crops again- on them I've drawn an arrow to show the direction of star travel. If the gap was due only to the lost exposure while the shutter was closed, then the gap should be perpendicular to the star motion. It's not.

a) from south of east, the star is rising into the night sky -there are about 17 gaps, so, this was about 18 shots. It's too regular to be 'twinkling'. Besides, a 60 second exposure should integrate all potential twinkles into the average star temperature color.

b) north of west, the star is setting towards the horizon

c) full image, crops are from the right and left side (this was just a fun image while out working- not one of my better images)








What it actually looks like to me is a scan line edge on a georeferenced satellite image caused by the satellite progressing along it's track while it is scanning.

JC

NeilL wrote: »

Where does your equation come from? A wildcard constant doesn't seem to me to be legit. Its value should be inferred from the requirements of the equation. What is that?

Yes, it's plausibly some kind of "scan" artifact, or sensor sampling rate, tho as I said I would have expected that to be much faster than the visible evidence.

The orientation of the gap lines suggests to me something analogous to a resultant vector. If so, the effect is being created by movement (fluxion-calculus), which doesn't seem to be in your equation, unless it's the constant. Rotation of the sky is a constant, and so is the relationship between that and the periodicity of twinkling averages (if that exists), or perhaps sensor samplings.

In my pic above, the movement of the camera together with the relationship of that to the cycling of the neons, seems to be the origin of the bands in the light. It is not smooth motion blur, which you might expect from sunlight. There are "beats", and beats are the regular cancelling/augmenting of data by data, a "resultant vector"-type value, as I would expect. Same kind of thing as the backwards rotating wagon wheels in the movies where there are clear beats (and possible orientation off the perpendicular to the axis of rotation).

Neil

I made up the equation, to try to clarify what I think. Of course, it should have said proportional to, instead of equals to, because there are scaling values in the first bit that I'm missing.

I think I want to add another variable:

F= focal length
T=time between exposures
A=angle away from the equator
G= size of the gap
L=Luminosity of star

C= unknown constant

G=T*F*(90/A)/L +C

The brighter the stars, the smaller the gap, all other things being equal, which could further suggest something in the encoding part of the process.

I didn't get everything you said, but a vector map of the gaps is a good idea, doing a quick and dirty one now.

NeilL

kolibri wrote: »

The brighter the stars, the smaller the gap

You are not convinced about the value of T, so why include it?

Above quote suggests signal-noise ratio. It also suggests a threshold, if the difference in size of gap is constant. Either way, brightness is irrelevant to the basic phenomenon ie the existence of beats. Therefore L should also be removed from the equation. It is the existence of the gaps which is to be demonstrated, and only tangentially their size. First get why the beats, then why they are what they are.

The equation needs a relationship between at least 2 constants of differing periodicities. Not T or L.

Neil

JC

NeilL wrote: »

You are not convinced about the value of T, so why include it?


Neil

What I'm trying to show, is that for longer times, say a minute if I had left the LENR on, then there would be a gap proportional to the time the shutter was closed. But as T=>0, the gap doesn't go to 0, it goes to some finite (non zero) size that is a product of some hardware feature, like the Bayer pixel pattern, or some software encoding function, unless the trail is soo bright that the values bleed over between exposures.

I wish I had a copy of some of my original startrails, that I processed without the screen-lighten method, because they show more clearly there, but i don't have them on this computer.

NeilL

I substantially edited my last post.

I think there are 2 questions here: why the gaps exist, and why the gaps have the properties they have. The answers might not be identical.

Unless you have images of star trails without gaps, and their absence is a function of T or L, then T and L are not part of the answer to why there are gaps.

The answer to that is the same as the answer to how to eliminate those gaps. Not T and not L. And maybe not other things in your equation.

Neil

kolibri wrote: »

What I'm trying to show, is that for longer times, say a minute if I had left the LENR on, then there would be a gap proportional to the time the shutter was closed. But as T=>0, the gap doesn't go to 0, it goes to some finite (non zero) size that is a product of some hardware feature, like the Bayer pixel pattern, or some software encoding function, unless the trail is soo bright that the values bleed over between exposures.

I wish I had a copy of some of my original startrails, that I processed without the screen-lighten method, because they show more clearly there, but i don't have them on this computer.

ziggy53

While the 1 second interval between the exposures would not account for this much gap, I think that any sharpening of each exposure could explain much of the effect. USM sharpening, in particular, can introduce contracted detail as it reforms the image edges.

Defer any sharpening until the image composite is finished. If anything, you should reduce sharpening to "0" during capture (JPG) or initial individual image processing (RAW).

JC

ziggy53 wrote: »

While the 1 second interval between the exposures would not account for this much gap, I think that any sharpening of each exposure could explain much of the effect. USM sharpening, in particular, can introduce contracted detail as it reforms the image edges.

Defer any sharpening until the image composite is finished. If anything, you should reduce sharpening to "0" during capture (JPG) or initial individual image processing (RAW).

Thanks for brainstorming this with me.

I shoot these in RAW.

Typical Post processing entails:

Decreasing exposure slightly (preparation for screen mode).
Raw impulse noise removal (stuck pixel removal)
Converting to 8 bit jpg (converting to 16 bit tiff didn't seem to make any difference-although I had to then convert them to 8 bit tiffs to stack them)
Screening subsequent images together.
Lighten-only stacking of all derivative image pairs.

This effect happens regardless of whether I do any additional raw noise removal, or regardless of the in-camera high ISO noise settings.

Here is a crop from 10 exposures I just re-did. Raw conversion included no noise reduction at all, no sharpening, no levels, no exposure variation. I combined them solely in lighten only mode, without the screen-lighten technique that reduces the gaps. You can see that the individual star shots have a rhombohedral shape- instead of a rectangle.

I'm not the only one that gets this effect- almost every webpage that offers tutorials on startrails talks about this, and they all say it's due to the time the shutter is closed, which I think is not a sufficient explanation. I'd love to get access to a fovean sensor camera to test with.

a) individual frame- east of north

b) 10 image stack, lighten only mode.


c) source image

NeilL

kolibri, as you know I think the starting point of your investigation must be a situation where the gaps do not happen. Could I ask:

? do they happen with film
? can they be eliminated with different camera+intervalometer configurations
? stack processing choices, eg blending mode
? aperture-diffusion
? lens fov
? direction of camera to the sky plane
? infrared

Neil

JC

NeilL wrote: »

kolibri, as you know I think the starting point of your investigation must be a situation where the gaps do not happen. Could I ask:

? do they happen with film
----not single long exposures


? can they be eliminated with different camera+intervalometer configurations

---

None that I have tried (XTI and 5D mII, several off brand intervalometers, all end up 1 second between exposures, at least for the 5D, it doesn't need additional write time between frames)

? stack processing choices, eg blending mode
---none that I have tried, and I haven't read anything else out there where people have found blending modes that completely eliminate them. Screening first then lighten only blending reduces them, doesn't eliminate them.

? aperture-refraction
---I shoot with the aperture closed about one stop, smaller apertures would require higher ISOs. I tried higher ISOs, up to ~800 , and I see other shots online with higher ISO's and they all have the gaps. I've never tried to significantly defocus the stars during shooting, I could try that

? lens fov
---- longer focal lengths changes the path length on the sensor between the gaps more than the width of the gap itself, but I think that there is also an increase in the gap width- but it' not completely proportional to the focal length, i.e, if I had an infinitely wide lens, i'd still see the gaps.

? direction of camera to the sky plane
----gap size increases with increasing angular distance away from north

? infrared
---I've never tried infrared star trails.

Neil

To sum up, yes, the root cause is the star motion, which increases relative to the sensor size at longer focal widths, and away from north. But, the characteristics of the gaps indicate some encoding issue, at least IMHO.

next time I'm out, I can try varying a bunch of the variables you've mentioned in a more systematic manner.

NeilL

kolibri wrote: »

To sum up, yes, the root cause is the star motion, which increases relative to the sensor size at longer focal widths, and away from north. But, the characteristics of the gaps indicate some encoding issue, at least IMHO.

next time I'm out, I can try varying a bunch of the variables you've mentioned in a more systematic manner.

Interesting.

It can't even be as simple as motion-FL-angle, since it doesn't happen with single exposure film, and it doesn't happen with "normal" motion blur.

Does it occur at mid-summer at the north/south poles? (!:D) I am thinking here also of contrast ratios.

You know I'm sure, that one cosmetic fix in post processing is to composite in a copy of the trails which have been rotated. But of course, those two sets of trails would not be perfectly congruent when offset because they should be in different parts of the sky. Makes me think that the progressive reorientation of the sky with respect to the sensor (a non-flat plane with a flat plane), perhaps additionally combined with the off periods between exposures, might produce a periodic altered/weakened sample/trace, as some parts of an already present trace are being overlaid with other traces, or those off periods (or other cancelling factor). IOW, the interaction of the sensor trace, which is a flat matrix, with subsequently differently oriented (in relation to earlier laid ones) trace matrices from a non-flat source, some of which are no-signal, as the sky moves. I imagine something like this situation would produce around pixel frequency beats which angle away from the perpendicular, and with a magnitude resembling a resultant vector, which is what is observed, if I am not mistaken. The calculation must need calculus-topology, which is what I would expect.

That would explain why gaps of this nature don't happen with film, because the grains are random, there are no gaps between them, and the exposures are summative only (not averaging as with an electronic sensor traces).

Neil

Sand

Kolibri:
May I ask a few additional questions about the gaps in your star trails.
Are the gaps aligned with the sensor pattern (the edge of the image)?
What is the approx. width of the startrails in pixels?
What is the width of the gaps in pixels?
What do the startrails look like where the trails are moving horizontal in the photo? At 45 degrees?
I am not sure if the answers will tell us anything, but it is an interesting effect.

NeilL

Sand wrote: »

Kolibri:
May I ask a few additional questions about the gaps in your star trails.
Are the gaps aligned with the sensor pattern (the edge of the image)?
What is the approx. width of the startrails in pixels?
What is the width of the gaps in pixels?
What do the startrails look like where the trails are moving horizontal in the photo? At 45 degrees?
I am not sure if the answers will tell us anything, but it is an interesting effect.

Looking just at kolibri's samples here, I would say that the gaps do not align with the edge of the image. Kolobri says the width of the gaps changes: "G=T*F*(90/A) +C"

The orientation of the gaps to the direction of star travel seems to change through an image. And the gaps are periodic, and the frequency seems to be constant throughout the image, and even perhaps across images (presumably if taken with the same lens, sensor, and camera settings).

Is it a kind of moire?

Neil

JC

I'm now leaning towards this being an effect introduced by the anti-aliasing filter. I now want to get my hands on a foveon sensor camera, and a camera with no anti-aliasing filter in front of the sensor to test.

NeilL

AA filter has been on my mind. Thought of it in association with the moire idea above. However an AA filter cuts frequencies above the resolution of the sensor in order to bring the system into line with the N-S sampling theorem.

Questions would be:

? are the input frequencies in time lapse night sky photographs more than +3x> the resolution of the sensor; what would be the source, since the bands are nowhere near such frequencies I think, and the AA filter should in any case have removed a moire-type (interference pattern) cause of them

? why with an AA filter do these bands not show up in non time lapse images

? since moire is *more* likely to show up without an AA filter, how could removing the AA filter remove the observed pattern

Neil

kolibri wrote: »

I'm now leaning towards this being an effect introduced by the anti-aliasing filter. I now want to get my hands on a foveon sensor camera, and a camera with no anti-aliasing filter in front of the sensor to test.