Subframe Selector Expression grading low-altitude photos as better than zenith photos?

brent1123

brent1123

Member
Hello all,

I have been experimenting with expressions in Subframe Selector (SS) to grade my images, as many others do, using a mix of star properties and overall SNR. The problem is PixInsight deems images closer to the horizon as having higher Weight, likely due to the higher average ADU value across the photo (whether due to air mass or LP signal) appearing like a brighter and higher SNR image.

I was experimenting with this expression which uses various properties of the subframe to hopefully generate a fair grade:

(15*(1-(FWHM-FWHMMin)/(FWHMMax-FWHMMin))
+ 20*(1-(Eccentricity-EccentricityMin)/(EccentricityMax-EccentricityMin))
+ 30*(1-(Median-MedianMin)/(MedianMax-MedianMin))
+ 15*(SNRWeight-SNRWeightMin)/(SNRWeightMax-SNRWeightMin)
+ 15*(Stars-StarsMin)/(StarsMax-StarsMin)
+ 5

However, due to the aforementioned issue with my overall image weight sliding upward as the target approaches the horizon (due to use of the SNRWeight property, which is removed in the expression below), I eventually found that the 'Noise' and 'NoiseRatio' values showed exactly the trend I was looking for - that the closer the target was to the altitude, the lower these values are. I edited this property into my expression to form this:


(10*(1-(FWHM-FWHMMin)/(FWHMMax-FWHMMin))
+ 10*(1-(Eccentricity-EccentricityMin)/(EccentricityMax-EccentricityMin))
+ 10*(1-(Median-MedianMin)/(MedianMax-MedianMin))
+ 10*(Stars-StarsMin)/(StarsMax-StarsMin))
+ 50*((Noise-NoiseMin)/(NoiseMax-NoiseMin))
+5

As I was editing the constants at the front of each line I eventually found that giving the Noise line an overwhelming amount of power would give my photos a grade which slowly dropped as the target approached the horizon, however the new problem was that the highest graded image never passed 65 or so, while the lowest was near 15. This problem persisted regardless of how high or low the constant values seemed to be.

__________________________________________
My questions on this topic:

- My usual method of using Subframe Selector involves using an expression like the first one above, but still manually going through the Eccentricity and FWHM and removing any subframes which have high values, thereby removing images with soft focus or tracking issues. Since I am essentially manually grading images before letting the SS process save the Weights onto each image, would I be better off abandoning the various properties of the expression and using only the Noise or NoiseRatio values for grading the images

- Related to this, would I then be better off skipping SS entirely (except for removing high Eccentricity/FWHM subs) and using the new NormalizeScaleGradient Script since it would effectively accomplish the same thing, as in judging my lease-gradient image as the best?

- Finally, regarding the fact that my highest image weights were capping out at 65 - how can I tell the expression editor to "set best image grade to 100" and then grade the rest on a curve? Is this even possible with how SS normally works?


Any help would be appreciated, thank you
 
mmmh, maybe this will help.... In SFS you load a number of images and one image has some cloud, then the image with cloud will have the highest SNR because it is brighter and not necessarily the best image. So using that image will skew the results. I prefer to use the script NormalizedScaleGradient which also lists the images with "Altitude" and you can then select the image with the highest "altitude" as the reference.

John
 
In the incoming version 1.8.8-10 of PixInsight we have two new image weighting algorithms tightly integrated in our preprocessing pipeline. This includes the ImageCalibration, Debayer, SubframeSelector and ImageIntegration tools. The current "noise evaluation" weighting option has been removed and replaced by the new methods. The new algorithms should solve these problems.
 
brent1123 said:
I have been experimenting with expressions in Subframe Selector (SS) to grade my images, as many others do, using a mix of star properties and overall SNR. The problem is PixInsight deems images closer to the horizon as having higher Weight, likely due to the higher average ADU value across the photo (whether due to air mass or LP signal) appearing like a brighter and higher SNR image.

I was experimenting with this expression which uses various properties of the subframe to hopefully generate a fair grade:

(15*(1-(FWHM-FWHMMin)/(FWHMMax-FWHMMin))
+ 20*(1-(Eccentricity-EccentricityMin)/(EccentricityMax-EccentricityMin))
+ 30*(1-(Median-MedianMin)/(MedianMax-MedianMin))
+ 15*(SNRWeight-SNRWeightMin)/(SNRWeightMax-SNRWeightMin)
+ 15*(Stars-StarsMin)/(StarsMax-StarsMin)
+ 5

However, due to the aforementioned issue with my overall image weight sliding upward as the target approaches the horizon (due to use of the SNRWeight property, which is removed in the expression below), I eventually found that the 'Noise' and 'NoiseRatio' values showed exactly the trend I was looking for - that the closer the target was to the altitude, the lower these values are. I edited this property into my expression to form this:


(10*(1-(FWHM-FWHMMin)/(FWHMMax-FWHMMin))
+ 10*(1-(Eccentricity-EccentricityMin)/(EccentricityMax-EccentricityMin))
+ 10*(1-(Median-MedianMin)/(MedianMax-MedianMin))
+ 10*(Stars-StarsMin)/(StarsMax-StarsMin))
+ 50*((Noise-NoiseMin)/(NoiseMax-NoiseMin))
+5

As I was editing the constants at the front of each line I eventually found that giving the Noise line an overwhelming amount of power would give my photos a grade which slowly dropped as the target approached the horizon, however the new problem was that the highest graded image never passed 65 or so, while the lowest was near 15. This problem persisted regardless of how high or low the constant values seemed to be.

__________________________________________
My questions on this topic:

- My usual method of using Subframe Selector involves using an expression like the first one above, but still manually going through the Eccentricity and FWHM and removing any subframes which have high values, thereby removing images with soft focus or tracking issues. Since I am essentially manually grading images before letting the SS process save the Weights onto each image, would I be better off abandoning the various properties of the expression and using only the Noise or NoiseRatio values for grading the images

- Related to this, would I then be better off skipping SS entirely (except for removing high Eccentricity/FWHM subs) and using the new NormalizeScaleGradient Script since it would effectively accomplish the same thing, as in judging my lease-gradient image as the best?

- Finally, regarding the fact that my highest image weights were capping out at 65 - how can I tell the expression editor to "set best image grade to 100" and then grade the rest on a curve? Is this even possible with how SS normally works?


Any help would be appreciated, thank you
Click to expand...
The resolution of faint extended objects is dominated by the signal to noise ratio. It is counter intuitive, but tracking errors and seeing problems usually don't have a detectable impact on these faint extended objects. The more frames you reject, the lower the resolution will be. It is a terrible shame to throw away signal that we worked so hard to obtain!

Tracking errors, scattered light and seeing problems do have a noticeable effect on stars. Bright point sources, such as stars, are a very harsh test. Fortunately, data rejection is often very good at getting rid of the bloat around these stars. If you are still not happy with the star resolution, I would recommend creating two stacks - one for the faint extended objects, and use the other to replace the stars.

Accurate image weights are also crucial to the final signal to noise ratio. I think the test you outlined is very well thought out. If a weighting scheme fails to cope with increasing air mass, or higher levels of sky brightness, then it clearly isn't working optimally.
 
Last edited:
You must log in or register to reply here.
Back
Top