Mure Denoise - Pixinsight

[Adam1234]

I've been playing with the MureDenoise script in Pixinsight today - spent ages trying to understand what values I needed in the gain (e/DN) and Gaussian Noise (DN) since I don't have the individual uncalibrated flats and darks anymore to use the DetectorSettings script. Anyway, after lots of research on the net and using the ASI1600 graphs I managed to calculate an estimate of these values (found out I had to convert the ASI1600 12-bit values in 16-bit - what a headache trying to understand!) FYI using Gain 200 I have calculated 0.03 e/DN for Gain and 46.70 for Gaussian Noise.

Anyway, now I've got that bit pretty much sorted I had a decent looking result come out of MD using the Variance Scale default of 1.000 (much better than the original image before MD anyway!). 

My question thought is this - I applied the MD with STF applied so I can see the result - great! I then close STF, and then decide I want another look at the image and apply STF again - why is then that applying a new STF makes the resulting image look horrible, but looks fine with the original STF? Is it something to do with a change in the background levels affecting the auto-stretch?

I'm guessing the way around this for the remainder of the linear processing would be to either keep the original STF applied and don't remove it, or create a process icon for the original STF. This seemed to work anyway. 

 Here's the original stacked image before MD (with STF), after MD with original STF and after MD but with new STF (all cropped).

Original before MD

 

After MD (Original STF)

 

MD (New STF)

 

 

Cheers

Adam

[wimvb]

Stf tries to apply a best histogram stretch based on black level, white level, and image statistics. Any noise reduction will change the numbers that are used for the stretch. But you can always change the stf  by opening the stf tool and moving the sliders. You may have to zoom in on the stf tool.

[david_taurus83]

Sometimes STF doesn't give the best stretch when you want to go non linear. Try stretching manually with the HT with small increments like you would with levels in Photoshop. It gives you finer control. I usually make copies and try STF, manual stretch, MaskedStretch and Arcsin stretch and see which I prefer.

[Adam1234]

Thanks, I've started using a combination of MaskedStretch and HT rather than relying on STF

[endless-sky]

Each noise reduction technique allows for a more aggressive stretch to be applied to the image. The autostretch will try to take advantage of this by pushing the image more, everytime you perform an action that reduces noise. If you actually like a stretch from a previous step, you can create an instance of it by dragging the triangle on the workspace. That way, if you accidentally perform a new autostretch and do not like the results, you can always go back.

I used to rely on autostretch almost entirely. Now I try to use iterative approaches, masked stretches, and play with masks and HistogramTransformation in a more manual, controlled way. I feel I can achieve better results, this way. The autostretch is, more often than not, too much, for my taste.