Image Stacking first test.

[Garbz]

I'm testing a different way of doing astro photography than my usual method. The problem with digicams is that unless you have specialised hardware pixels die and the edges of the frame start to bleed pink because of heat. My last attempt at a 2 hour exposure, along with a second 2 hour black reference exposure to eliminate the pink noise, and about an hour of post processing to remove dead pixels was average. Not the least of the problem was the light pollution which is nasty here!.

Anyway I tried the stacking technique yesterday:

This image was made of 60 images at f/3.5 ISO100 for 30seconds. The first 1 images were stacked and added together, and then the dust and scratches tool in photoshop was used to remove the stars. This gives the reference brightness of the foreground and background.

The reference image was then stacked with all 60 original images using a method that selects only the brightest value for each pixel on the final. The result is that 60 exposures of the stars were super imposed over the original background giving star trails with NO noise.

If you look closely enough you can count the number of frames taken thanks to a typo by me which set the exposure time to 30 seconds and the timelapse interval to 45. I will try again tonight to get something smooth and perhaps even something more interesting in the frame

 

[astrostu]

I'm not quite sure I understand your technique ... lemme see if I got it:

(1) You took "the first 1 images" (so, first how many?) and averaged them together, then removed hot pixels and the stars in PS.

(2) You took the actual time-lapse photos and added them together via a MAX() as opposed to AVG() routine.

(3) Then you did the same thing with your final two images to get the stars back onto the background.

Is that it?

It's an interesting idea, but I would caution you against saying that there was "NO noise." There is always a statistical uncertainty - governed by Poisson statistics mainly - in the actual photon count recorded by each pixel. Especially since you didn't do any dark-subtraction - which takes care of the additional Gaussian noise - the idea that there is "NO noise" is very misleading and quite incorrect.

You also might want to clone out the airplane.

 

[PhotoPhoenix]

uhm. you're a complete genious by the way. dude, i've been trying to do star trails for so long now with my 30D but it's just not made for these types of things and the noise gets so bad after 5minutes. it's not worth it. this is an awesome technique. i can't wait to try this out. awesome dude, awesome. the photo looks verrry very clean. you should really get someone to publish this technique.

 

[Garbz]

Ok maybe I shouldn't have capitalised the NO then . It's true that it is almost impossible to get an image with NO noise. But ignoring the physics, statistics and going back to pure photography I would define noise as something unwanted or distracting in an image. If you've every tried taking a 1h DSLR picture of something with an average background complete with hotspots, gaussian noise and bleeding of the frame edges, you would have to agree by comparison the above does in fact have no noise

Sorry I didn't explain it correctly. I took 60 images at the same settings. That's it. The problem was the background was way underexposed. So I combined 10 images (sorry the 1 is a typo) using ADD(). Then I went in with photoshop to remove any star trails, giving a foreground and an detail-less sky.

I then combined the resulting 61 images using MAX() to produce the above.

I repeated this experiment tonight to try and make an actual image:

This time I bumped the ISO up to 160 to get the brightness rather than stacking images. This shows more stars in the sky and noise was still very acceptable. The result is 200 images stacked using MAX() none of the other mucking about to get the right background.

It's interesting to note that I set my camera to 30 second exposures, but set the timelapse to 33 seconds. Yet almost the same instance my exposures finished the next photo was taken. Either the timelapse or the camera's exposure metre is out either by accident or on purpose. No mirror lockup or anything was used so in theory there should only be about 30seconds exposure, some 10ms mirror delay, and then a 3 second gap, but no...

Also on the original frame despite being exposed for 30 seconds and what seems like only a 1 second gap in between frames you could still see the individual stars on the left of the frame away from the southern point.

 

[shorty6049]

photophoenix- i'm pretty sure this has already been published, becase i'd heard of it before i read this post, but that looks very nice garbz! its too bad that you cant see many stars, i would suggest that you use a higher ISO to get more stars in your shots . the second one looks better though. I havent had the patience to do a star trail using the stacking method..., i end up getting bored and going inside. but astrophotography still is fun, isnt it?!

also, you were talking about the 2 hour dark frame , my sony alpha has a noise reduction feature in it that does exactly that (or at least i'm PRETTY sure thats whats going on) if the NR is turned on, it takes twice as long to take a picture, because after the exposure is over, the "processing" sign comes up for the same length as the exposure was. kind of neat but as you said, its just OK, not great, dont know if other cameras do this or not?

 

[Garbz]

you should really get someone to publish this technique.

Sorry I missed this the first time around. This is actually a well known technique in astrophotography, although it is not normally used for star trails. Often it is done to minimise noise in long exposures by taking long photos of the sky with some kind of tracking system, and then averaging each pixel. Since the average of any Gaussian distribution with a significantly large enough sample size is 0 noise is reduced and celestial bodies remain.

The other way it is used is by adding each frame. This allows you to bring more detail out of parts of the frame that are even completely black. This is great if you can't align the telescope properly but you still want to bring out details that would normally require an exposure of several minutes. I will try this technique next new moon in the bush somewhere.

Shorty as you can see in the second photo with the higher ISO, I have effectively reached the limit as to how many stars I can do this with, without leaving the house. I agree, it's incredibly boring so I setup the camera and go back inside. Light pollution here probably won't allow this to get any better than it is currently. If I feel dedicated enough I may take this into the bush sometime.

Every DSLR has that form of noise reduction feature. The reason I left it to post processing though, this method it would have resulted in pretty huge gaps in the star trails (30s exposure, 30s NR). My batteries would only last a total of 2 hours continuous, or a bit more with these individual frames, so it halves my effective time. Noise distribution is similar, and hot pixels are the same for a given exposure time so only 1 photo was needed for all frames. Also this method of NR doesn't work for pictures longer than say 30minutes or so. Since the pink bleeding is caused by the CCD heating up if it takes a 1h exposure, and then straight away the NR image the CCDs would be heated differently during each frame and thus subtracting the pink bleed would not work as well. It helps a lot for exposures of a few minutes though.

 

[shorty6049]

oh, yeah, sorry i didnt know wether that was the way all cameras did the noise thing or not, i have a lot of light pollution in my area and one way i was able to get a lot of stars to show up was with a moderately high ISO and an aperture of 1.8 or something close, but yeah, LP does come into play then as well

 

[PhotoPhoenix]

what program is MAX?? how would i go about doing this with Photoshop CS3? layer masking? seems like that might take forever...

 

[astrostu]

what program is MAX?? how would i go about doing this with Photoshop CS3? layer masking? seems like that might take forever...

By "MAX" I was referring to the method of image addition -- taking the maximum value at a given pixel across all the images.

I haven't explored it in CS3, but I know that CS2 does not do easy image arithmetic.

 

[Garbz]

And I was playing along with the analogy. Photoshop as far as I know only does a mathematical subtraction and multiplication out of the basic arithmetic set. But search for image stacker on google and you'll come up with many programs capable of doing this.