Lessons from Thierry

6/24/2025. Sometimes, when better astrophotographers than I post their work, I find it inspiring. Then there's Thierry Legault. I look at his solar photos, and for just a fraction of a second I think, "Wow! Look what's possible!" And then I look at mine and think, "Why bother?" Here's a recent example made with a C8:

https://solarchatforum.com/download/file.php?id=112888

Can you view that link without signing in on the SolarChat Forum? It's a killer.

Here are some earlier examples made with a C11 on Thierry's site:

http://www.astrophoto.fr/C11HAT_2015_2016.html

These results have sent me looking for the errors of my ways. He's using about twice the aperture I am, but at first glance it looks as if he's got about ten times the detail. And those tones! And that complete lack of noise! How in the ever-lovin' hell?

I pulled his recent image into Photoshop and started degrading it to see just how it compares to my best efforts. It's actually about 3x better resolution, but the fact that it is essentially noise-free with a full range of tones makes it much better than that would imply. The resolution is largely down to aperture (C8) and very good air (and expert stacking, of course). It occurs to me that I'm doing exactly nothing to mitigate noise in my raw solar files despite pushing them much harder than I push nocturnal images for which I take great pains to reduce noise. Exposure times are orders of magnitude shorter (3ms vs 180s or more). In each case, I'm averaging hundreds of frames, and I really thought noise was under control in the daytime images by virtue of stacking very short exposures. But it's clearly not that simple.

Thing is, not all noise comes from long exposures. Some is inherent in the sensor, especially when the sensor is operating at 125F or more rather than zero or below. Plus, in processing these, I'm leaning hard on the data trying to pull up exactly the kind of high-frequency details that would make noise visible, too.

My best results today came from using a 1600x1300 ROI and no telecompressor. If I want fine detail, I must have image scale. If the conditions don't allow it, so it goes, but I'm not going to get fine detail without imaging at a generous scale.

I think I will work on cleaning up noise first and then see what issues are mitigated and which aggravated.

Am I introducing noise by using a flat made up of 50 frames (like the captures, acquired at high gain)? Will dropping gain substantially and exposing longer be a good trade off, at least when the seeing is decent? That should clean up both the capture and the flat. Is there need and a practical way to apply a dark frame to these clips? Using gain to "fill in the histogram" has been my usual practice, but perhaps I should abandon 2-3ms exposures and use 10ms or more. Should I try 8-bit captures (if they permit faster frame rates)? I need to try some very different settings compared to the ones I explored today (expose left, right, for high dynamic range, for contrast). I routinely expose to the left in an effort to freeze seeing, but that means I am having to work with noisy, dark data. Today, I also tried selecting thousand-frame (15-20 second) subsets from five thousand frame clips when seeing was good just in case motion on the Sun was an issue (it wasn't).

Next time out: try very low gain, gamma off, 8-bit captures, AVI, SER. Even if something else suffers, noise might be greatly improved. I'll take on one addressable issue at a time.

 

6/25/2025. No, amazingly, 8-bit captures are not faster. I grabbed several anyway, since 8-bit AVI's are smaller, will transfer faster, and are just generally less awkward to play with than 16-bit SER files. The sky is hazy, so exposures ran longish anyway. My usual 300-350 gain required almost 5ms exposures. At 100 gain, exposures were around 40ms. And at 231 gain ("unity" for this camera), about 12ms. But today we have working on noise. I didn't use flats today, because I didn't want to keep refocussing between every 2,000 frame "snapshot."

Also, amazingly, a stack of frames exposed at 44ms can be pretty damn sharp. I've been trying to keep exposures short (1-3ms) to "freeze seeing" so that stacking can work its magic, but the price of doing that is high gain and that means noise. You do get a brighter image by not applying a gamma correction, and you can get that contrast back in post, so that's something to keep in mind.

And another thing: it's surprisingly hard to tell an 8-bit clip from a bit shot in nominal 16-bit mode (which is really 12 or 14 with this outfit). They might even be better (because I suspect the 8-bit clips are less likely to capture pattern noise).

On the SolarChatForum, Thierry himself answered a quick question about methods posed and posted by RSFOTO a few days ago under the title "Freezing the seeing." Unless his data is perfect, I suspect much has been elided. But here you go:

Here is how I process.

• raw frames registration and stacking with Autostakkert (25 to 50 frames at gain 0, more frames at higher gain)
• sharpening (simple unsharp mask generally)
• use of PS function "Shadows & Highlights" (in "more options" mode), with adjustment of the three shadows parameters, midtones contrast and black and white clips, to reach a satisfactory result where the disk gets good contrast but not overexposed, with prominences highlighted in a background which is not completely black
• final adjustment of levels and gamma and colorization if needed

6/26/2025. As I was dozing off last night (no, seriously, I am that much into this stuff), it occured to me to try 2x2 binning to see if it would produce a) a better match to the available resolution and b) if it would reduce noise, including pattern noise that sometimes afflicts the 178MM and c) if it would let me use faster exposures at lower gain. It does do (c), seems to do (b), but (a) is still up for grabs. Using only a quarter as many pixels for my usual ROI selection is just not a lot of pixels.

Among today's best results was an 8-bit clip at gain 231, no binning, exposed for 23ms, at something like gamma 15. The sky was hazy, so exposures ran long again, and I wanted the extra contrast an aggressive gamma setting gave me. Once again, I tried a ton of variations -- enough that they're essentially blind trials -- and I haven't done the same systematic crop and close examination that I did with yesterday's take. All in good time. Today, FWIW, I did do fresh flats for most clips. And I batch processed the clips at native resolution and at my usual 3x drizzle.

For now: keep the gain low, don't be afraid of longer exposures, and not binning works better than 2x2 binning (except, just maybe, for full-chip imaging where there are plenty of pixels to work with). More conclusions and examples and ruminations to come.

 

6/28/2025. I have been misled by a metaphor! Once could be happenstance, twice could be coincidence, but the same result three sessions in a row begins to suggest that something needs to be paid attention to. The magic settings for my kit seem to be: no binning, 8-bit capture, gain 231, uncompressed focal length, and exposures on the order of 25-45ms, whatever the sky demands to put the peak of the histogram in somewhere in the middle of the graph. Go ahead and drop gamma to 12-30 to get satisfying, visible detail, but it doesn't really seem to matter very much to the final product. Focus like crazy on enlarged sections of the field displayed at low gamma. And use flats.

The usual locution about using video to "freeze seeing" does not really mean what I've thought it did for years now. It's not about using short exposures to get sharp detail from images in rapid motion due to boiling air. It means using video to capture enough frames that a useful number do, in fact, capture parts of the subject when they are not moving. So, nevermind trying to use exposures in the millisecond range. To do that, you need a lot of gain, and gain means noise, and to cope with that, you need a lot of frames -- maybe more than is practical with small glass and a subject which does, in fact, move. By using relatively low gain and longer exposures, I'm seeing a lot less noise and can get by on much shorter stacks (75 frames from 1,000 to 2,000 frame clips rather than a few hundred frames from 5,000 frame clips). Detail and tone are significantly better under this new paradigm.

Selecting 16-bit mode (which with my kit actually yields 12-14 bit data) sounds like a good plan, but with the ASI178MM, it really makes capturing the camera's notorious pattern noise much more likely. Oddly enough, so does stacking with 3x drizzle. Stick to native resolution unless there's a good reason not to, and use 8-bit mode in the middle of the histogram. If an ASI174MM comes up on Astromart at the right price, you can revisit that advice.

Third session's the charm:

You bet you can click it.
Yes, I know I am still pushing the data too hard.
Processing and acquisition details will follow after another session or two.

Question: can I go to lower gain and introduce an offset to shove the histogram to the right ("burying the read noise" in the vernacular)? And is that a good idea even if I can? I'm only using 90-120 levels in any given clip (the limb is different), so maybe this makes sense?

While waiting to find the answers to some of those questions, here's something from my fourth session with this new approach. Yeah, something works:

Clicking encouraged.
So maybe I wasn't pushing too hard; maybe I just needed better seeing.
This works well and consistently enough that I need to pull some notes together.
On the next page. Soon.

 

 

 

:: top ::