Superresolution Lunacy

Ever since my original moonshot of Longs Peak, I've been itching to go back. I want a bigger print of that shot than the file quality supports - caused primarily by the rather poor lens I used on the original trip. I also wanted a repeat with different light.

I left Denver at 02:00 last Friday morning, figuring that there would be a beaten trail to the summit of the Twin Sisters, that 2.5 hours should be plenty for me to gain 2700'. I figured wrong. What should have been a pleasant moonlit hike turned into a brutal time trial against moon and mountain. I had to break trail up the last 800', through deep soft snow with horrific breakable crust. Bad on the way up, it got even worse on the far side of Twin Sister East. In the end, I didn't make it to my intended position. When the moon looked to be about 6 minutes away from setting I had to stop where I was and start shooting, even though it wasn't the composition I wanted.

I've recently been  intrigued by super-resolution composites. The idea is to shoot a burst of images at the same composition and settings, align them, up-res, and average the result. Ideally, small camera movements allow the sensor to sample different scene elements in each image, and that average result can have a resolution well above that native to the sensor. Just as important, if not more so, the averaging process should reduce noise by 1/sqrt(N), N being the number of frames. With 16 frames That should amount to a 2-stop increase in acceptable dynamic range by suppressing shadow noise. 

The two images above show the result. The first is a single frame, and the second is the super-resolution composite. Both have had exactly the same sharpening and noise reduction applied in Lightroom. The three finished images posted here are all super-resolution composites of 10 images, all shot hand held. I settled on 10 frames, as that appears to be about the maximum my computer can handle in a reasonable timeframe. I'm quite pleased with how good the result is with little in the way of manual fiddling in photoshop. There is a very marked increase in resolved detail in the composite, though it is difficult to know if this is primarily due to simple noise reduction or the super-resolution effect. Either way, it allowed me to shoot images hand-held that would have absolutely required a tripod without this technique. In fact, I suspect that even using a tripod I could not have achieved as a good a result, since the exposure density in the shadows could not have been increased without clipping the highlights in the moon. One could certainly use an exposure bracketing technique to achieve a similar result, but the ease of shooting and processing with this technique seems superior for my purposes.

I'm still not sure I have the shot I want hanging on our wall, so another moonlight stroll will be in order at some point. Next time I'll leave at 01:00 and make myself a cup of tea while I wait for moonset at my intended location.

Update: 
I can do more frames if I get my operations in the right order. Best practice seems to be:
1. Load images as layers
2. Align images into even groups in batches of less than 10 (auto align, no distortion correction.
3. Convert to smart object.
4. Up-res.
5. Set stack mode: mean or median (I've not found much of any difference, though median should be more tolerant if there is a single mis-aligned frame).
6. Rasterize and delete black shaded areas created where there was no image data.
7. Repeat step 2, 3, and 5 to combine the groups of 10.
8. Rasterize.

Of course, for these moon photos I had to cut out the moon from each frame and align mountain and moon separately, as the 1-2 seconds it takes to capture a burst of 30 images is enough time for the moon to move by several pixels. Using 30 images instead of 10 leads to a noticeable improvement in quality, particularly in the sky smoothness.

Full size files are here: