Stacking, Filtering, and a Lunar Preview
Took the scope and webcam out for a spin again last night. The moon is just past first quarter, which means at this time of year that it’s very low in the sky (suspiciously like what the sun will be in three months), and hard to get a good look at. I had to set up in an unusual location to get a shot that was unobstructed by trees and houses, but once I found the spot I got in a good hour of moon imaging. Mmmm, heaven.
I took forty-four individual videos in all, about 100 frames each, for a mosaic covering the entire illuminated surface of the satellite. The unregistered shareware I was trying out last night decided for some of the shots that 100 frames was too many, and instead took around a dozen. I guess I should appreciate the help, since I’m new at this, but gee whiz. This does affect results. The observant will notice some parts of the moon will be fuzzier than others when the mosaic is finally available.
That bug did expose some interesting, if expected, behavior in the digital imaging process. The whole point of image alignment and stacking is that when multiple frames of the same image are stacked, the amplitude of the images’ noise component is reduced. The signal portion of the image (that is, what the subject actually looks like, not speckles introduced by the low-cost camera electronics or waviness introduced by atmospheric distortion) then becomes stronger in relation to the noise. The more frames you stack, the better your signal-to-noise ratio, increasing in proportion to the square root of the number of stacked images. One hundred frames is therefore about three times better than ten frames.
After stacking, the most excellent Registax provides an image processing mode that allows the user to adjust the parameters of a “wavelet” filter that can bring out extra detail in the subject. You can only go so far with this before the filtering process introduces undesirable artifacts, not the least of which is accentuated high-frequency noise. You can turn up the filter to get very crisp-looking lunar mountain ranges, but for your trouble you might get very speckly-looking maria, which should be smooth and flat.
On the 100-frame shots, the signal-to-noise ratio is such that the filter can bring out a decent level of detail before the artifacts start to dominate, even on the fully illuminated parts of the moon that are traditionally very low in contrast. On the bonus ten-frame shots afforded me by this wonderful shareware tool, we had roughly 70% less joy. I was only able to turn up the filters ever so slightly before the noise started to pop right out of the picture. The resulting pictures don’t look any noisier due to having been hand-tuned with an eye towards this effect, but they are of necessity less detailed because of it.
I see that look on your face. Maybe I should explain.
First, look at the following three images (click for full 640×480 glory; new browser window will open). The first is a single, raw frame acquired from the webcam; the second is the result of Registax aligning and stacking 100 such frames; and the third is the output of a hand-tuned wavelet filter in the same program. (You have to click and drag the sliders–sliders!–yourself. This filtering business is very technical.)
Hopefully you see that the raw frame is very fuzzy in detailed areas and uneven in the smooth areas. The stacked image is also very fuzzy, but edges and flat areas appear smooth and uniform. The filtered image then brings the detail back–detail which has been hitherto invisible.
The discerning eye will spot some differences in this next set. Here is where AstroVideo (no link; I’m mad at them) decided ten frames would be better than the requested 100:
Numbers don’t lie. Those few shots that got shorted will be a little fuzzier than the rest in the final mosaic, but they shouldn’t diminish the overall presentation too badly.
Incidentally, I’ve spent some time in Photoshop messing with the compositing process. It’s a bit more time consuming than I had hoped, as the shutter speed and sensor gain settings had to be varied from shot to shot to maximize contrast, requiring me to adjust brightness manually on each of the 44 tiles as I align them. The final picture will be somewhere in the neighborhood of 3200×4800 pixels, and should appear to reveal a startling level of detail when scaled down to manageable sizes. I’ll be sure to share it as soon as it’s done.
-
- Posted by Tim on Friday, September 5, 2003 at 7:39:27 pm PDT
- Category: Miscellany
- Comment RSS Feed
