@49616E: Any chance you could outline your specific LDA method? I converted a copy of my original working file (just the aligned raw images, no other adjustments) to 32 bit, and set all 30 frames to Linear Dodge Add. The results were far noisier than the mean/median approach, and I seem to be having problems tuning the exposure/levels.
Here are my results, using the various methods I know.
1. The original method I used to use was something I found years ago on a basic astro photography stating tutorial. I used this to a degree in my original post, I call it the gradient ramp method, where you successively increase opacity of each layer from the top to the bottom of the layers list. It is simple, moderately effective, but breaks down when you start to tweak the results with adjustment layers and whatnot.
2. The 32-bit float (HDR) LDA (Linear Dodge Add). First time I've tried this with 32-bit. With 16-bit, you can't stack too many frames before the whole image just becomes white. As such, you can't really remove a lot of noise. Seems moving to 32-bit improves your options...but I may be doing something wrong, as the results (without any other tweaks) seem to be the worst of the three.
3. The median method, where you stack into a smart object then set the stacking mode to Median or Mean. PHENOMENAL noise reduction, especially when you stack a lot of frames. It can almost entirely eliminate noise. Your editing latitude goes through the roof. If you stack too many frames into a smart object, you'll start to lose color fidelity (in the example below, all 30 frames were stacked in one smart object, vs. my original post, where I created batches that were stacked then gradient ramped.) Original Light Frame (1 frame to show noise):Gradient Ramp Method (30 frames):Linear Dodge Add Method (30 frames):Median/Mean Method (30 frames):
I tweaked the above 30-frame stacks using the LDA and Mean methods. Assuming I did the LDA method correctly, once I started dropping on adjustment layers, there was a marked difference in editing latitude between LDA and Mean. Mean had far more before
noise started to exhibit. It was more forgiving and flexible, allowing fine tuning to tighter ranges of tones, without resulting in harsh gradients or transitions. I was able to maintain the black point better with Median than LDA (which ended up a little more red, but pushing it too far resulted in more noise.) Median preserved more subtle and dimmer tones in the outer regions of the nebula, where as LDA clipped more of the outer regions to the noise floor.
(Again, this is assuming I used the LDA method correctly):Fine Tuned LDA:Fine Tuned Median: