Recently I've been reminded that a small amount of very good light frames produce a better image than a large amount of bad light frames (this is related to my M1 expansion project which I hopefully will be able to continue next month)

So I've tried to integrate the best frames from 72 hours of rather good data, which my scope gathered last year for the NGC 2841 galaxy. This image uses 320 best frames out of total 1438, which corresponds to 16 hours. The frames have lowest FWHM and highest number of stars. I think that this allowed to reveal a lot of fine details in the galaxy. Revision C is an animation which compares 1438 frames with 320 best frames of the integrated G channels. Both stacks used frame weight from PixInsight's WBPP script with the following Subframe Weighting: FWHM 30, Eccentricity 40, SNR 20, Number of stars 0 and Pedestal 10.

FWHM of the integrated image without deconvolution is 1.95 arcsec. The image scale is 0.435 arcsec/pixel.

Color calibration was done with AutoColor script in PixInsight.

Thanks for looking,
Vitali

Update 2022/11/07:

While discussing the result @Robert Shepherd  brought up the idea about using all frames for faint areas of an image and small number of best frames for sharper details in the bright areas. I've decided to try blending to combine the results of all frames (LowResolution) and best frames (HighResolution) integrations.

The idea is to use alpha blending (Target = a * HighResolution + (1 - a) * LowResolution) where alpha factor (a) depends on the brightness of a pixel. So bright areas of the combined image use pixels from best frames integration and faint areas use pixels from all frames integration.

Logistic function was used to compute alpha factor:
a = 1 / (1 + exp(-k * (x - x0))),
x0 is the threshold between high and low resolution images. Target pixels with brightness below x0 use pixel values from low resolution image, above x0 - high resolution and around x0 - a mix of two.

So I've opened both all frames (Ilowres) and best frames (Ihighres) integration images in PixInsight and used StarAlignment to register them to each other (just to be on safe side). Then I've used LinearFit on Ihighres with Ilowres as reference image.

These images were combined using the following PixInsight's PixelMath expression:
A = 1 / (1 + exp(-10000 * (Ihighres - 1.4 * median(Ihighres))));
Ihighres * A + Ilowres * (1 - A)

Where

Ihighres is the best frames integrated image.

Ilowres is the all frames integrated image.

k = 10000 defines how wide is the transition area between low and high resolution parts. Low values of k will make the transition smoother and produce less detailed result. Higher values of k may lead to artifacts around bright objects (stars, etc).

x0 = 1.4 * median(Ihighres) defines the pixel brighness where the transition occurs.

I've chose these "magic" numbers for this particular image after some experimenting.

Revision D shows the low resolution image (all frames), followed by blended image and the high resolution image (best frames).

Revision E shows autostretched images. Faint areas are more pronounced on the combined image compared to best frames integration.

Revision F is a quickly processed blended image.

I hope that this experiment may be helpful for some of you

Clear skies,
Vitali