Histogram "drop" after L combination <- problem or not?

[pfile]

what do the calibrated subs look like statistics-wise?

what happens if they are not flat-calibrated?

trying the older darks/bias seems like a good test.

rob

[Greg Schwimer]

Calibrated, registered examples below. I'll have to try running with the old calibration frames tonight or tomorrow. Kids....

Code: [Select]

RED
count (%)   99.79070
count (px)  16166093
mean        163.397
median      162.597
avgDev      14.251
MAD         11.485
minimum     45.388
maximum     25444.635

GREEN
count (%)   99.70872
count (px)  16152813
mean        280.181
median      279.210
avgDev      17.487
MAD         14.159
minimum     42.764
maximum     31018.546


BLUE
count (%)   99.80014
count (px)  16167623
mean        280.452
median      279.562
avgDev      17.444
MAD         14.197
minimum     53.065
maximum     18040.125


[pfile]

so then i guess the question is if the sky is so dark there that it was normal for the darks to remove almost all of the signal in the background?

rob

[Greg Schwimer]

I'm certainly in no dark sky. In fact, I'm pretty sure this is a white zone these days. Looking at the bias vs. dark value, wouldn't it be the bias that is removing most of it?

My prior master dark was bias subtracted. The new one was not so it's difficult to compare cleanly. I subtracted the master bias from the new one and they're quite close though.

To be sure, I bias subtracted the new dark frames then integrated with the same settings I've been using forever. Here are the stats for each of the master frames:

Code: [Select]

Old Bias
count (%)   100.00000
count (px)  16200000
mean        1013.221
median      1012.980
avgDev      1.485
MAD         1.152
minimum     1005.726
maximum     1067.650

New Bias
count (%)   100.00000
count (px)  16200000
mean        1030.969
median      1030.680
avgDev      1.660
MAD         1.280
minimum     1023.099
maximum     1087.980

Old Dark
count (%)   100.00000
count (px)  16200000
mean        11.185
median      10.806
avgDev      1.583
MAD         1.025
minimum     3.044
maximum     53785.250

New Dark
count (%)   100.00000
count (px)  16200000
mean        9.865
median      9.448
avgDev      2.162
MAD         1.517
minimum     0.580
maximum     54113.272


I don't think these differ enough to draw concern. Agree?

This does get me thinking though - did I expose long enough to get the background sufficiently above the bias and dark signals? The calibrated subs appear to be well enough above zero, but I'm not sure just how far above zero they really ought to be. I never had to look before.

Think it's still worth trying calibration with the old bias & dark?

[pfile]

sorry - i was thinking of an uncalibrated dark; its true that the levels in a master bias will be >> a calibrated master dark.

so yeah, probably there's no difference between your old and new calibration frames.

anyway, for comparison, with my STT-8300M, the mean bias level is around 950ADU and the mean signal in an uncalibrated 1800s dark is around 988ADU. in an uncalibrated 600s B frame (using an f/5.5 telescope in a bortle red zone) the background ADU is around 6900ADU (min around 6200ADU). so assuming perfect dark scaling of 0.33, the mean backround would be around 5940 after bias/dark calibration.

i can't really analyze your images as they've been flat calibrated so it's hard to be sure what the min value was just based on the sky. however the background ADU counts are in the ~180 range for the R frame and ~200 for the L frame. not that there's anything wrong with that - as long as you didn't clip any pixels during calibration then in theory you're good - but i think it probably means you could expose longer. those low-value pixels are prone to clipping when you stretch, which i think accounts for the weird histogram readout behavior.

are your lights super-short?

rob

[Greg Schwimer]

Not super short at all. 180s. I normally go for 300s, but I get clipped stars in some cases. I wanted pristine stars in this case, so I found an exposure that kept me there for each of the RGB filters.

If you want to see the unprocessed data and the calibration frames it's all right here:

https://www.dropbox.com/sh/7gu750oybp1asnz/AADlBlaubz1q0EqvnEDr4hyra?dl=0

I don't think I'm doing anything wrong in the preprocessing or processing part.

Ugh - just running through the exact same process as I started with (re-using process icons so 100% repeatability) and the drop doesn't happen now. Maybe I'm tired, but it's sure odd that the behavior is different. I wonder if there is a bug at work?

[pfile]

is it possible the dark got double-calibrated by ticking "optimize" in ImageCalibration during light calibration?

rob

[Greg Schwimer]

is it possible the dark got double-calibrated by ticking "optimize" in ImageCalibration during light calibration?

Nope. I used BPP in this case to create the master bias, dark, and flat frames and to calibrate the lights. The lights were manually registered and integrated.

And like last night, I can't seem to reproduce the "problem" if you can even call it that. I started from the initial RGB and synth L frames and took them all the way up through a stretch and combined. The histogram looks fine now. No idea what is different aside from maybe the STF parameters I used for the stretch.

[pfile]

i suppose given that rick did not have any problems it's possible that the 2nd time thru you didn't clip those low-valued pixels and so the spike of 0 valued pixels never appeared.

rob

[Greg Schwimer]

It must've been something with that. The only think I can think of is that previously I did have an amount of shadow clipping - about 0.05%. I mostly eliminated that.

Thanks for the assistance and brain storming!  Here's a result.

[RickS]

Not too shabby, Greg!

[jkmorse]

Greg,

Great image but one tweak you might try.  On my computer it looks like you have some blue running in the background, particularly towards the bottom.  One way to fix that is to clone your image, convert the clone to grey scale, invert it and then stretch to protect everything but the background.  Then apply that mask to your main image.  Once you are working on the background only, use Curves Transform to desaturate the background.  It will turn the background a consistent grey.

Best,

Jim