Large-scale pixel rejection maps?

[gregwjones]

I am trying to remove some stubborn sat trails in my stack of 100 subframes. I have been looking at the rejection_high image that is available after I run the ImageIntegration process and it does show significant rejection around this sat trail, but the stacked image still has some trailing. So, I enabled the "Reject high large-scale structures" and re-ran ImageIntegration with "Generation rejection maps" checked.

The process console indicated that it generated the high large-scale rejection map, but no such image was available when the ImageIntegration process completed.

Is the high large-scale rejection map included in the regular rejection_high image, or should it have generated a separate image for the large-scale rejection map?


I am using the new ESD rejection algorithm in PI 1.8.8. Linear-fit rejection also did not remove this streak.
 
Regards,
Greg

[pfile]

i think the problem here is the sat trail has some kind of gaussian distribution in cross section, so the outlying parts of the trail don't look like strong outliers. if you search the forum (may be easier with google using site:pixinsight.com) there are some posts describing how to use pixelmath to paint over the trail with truly outlier data (0 or 1). unfortunately it's a pain because of course you have to compute the endpoints by hand, there's no way to draw the trail, and of course it has to be done to each sub that has problems.

who knows with starlink going up to 10s of thousands of satellites maybe we're going to have to develop better tools for this problem.

rob

[gregwjones]

pfile, thank you for the reply, but my main question is:

"Is the high large-scale rejection map included in the regular rejection_high image, or should it have generated a separate image for the large-scale rejection map?"


I am looking for the large-scale rejection map image so I can see what the large-scale pixel rejection function is doing.

Regards,
Greg

[pfile]

well, i can't remember offhand. i'd have to try it.

i guess unspoken in my reply was that i wonder if the large-scale rejection will be able to detect the sat trail as a large-scale feature. i sort of remember when it was introduced that it might be good for rejecting large features like uncorrected dust donuts, etc.

rob

[gregwjones]

While I am trying to learn how to use the large-scale pixle reduction function I am also trying to remove this problematic sat trail. This thread indicated that some large plane/sat trails could be removed using large-scale pixel rejection. https://pixinsight.com/forum/index.php?topic=11067.0

This problem trail is in the M42 area that I collected over two night separated by 4 days.

 I thought that with 100 frames that all plane or sat trails would be rejected. But, they were not.  I found that this streak was caused by a slow moving object that also reversed direction. You can see this in the attached avi file. (you will need to rename the .txt file to .avi) This avi file was created from within the blink tool. I verified that all files loaded into the blink tool are in chronological order. I recently changed my filenaming convention in SGP so that the files would be listed in chronological order.

Regards,
Greg

[ngc1535]

Whoah... this conversation might be veering in a strange direction!

When you use "Large Scale Rejection" you are basically generating an image of rejected pixels and then... using the same kind of multiscale analysis found throughout PI, you find structures that will benefit from increasing their size (thereby including neighboring pixels and getting a better rejection). Although there is more math, I like to imagine this as a "growth" of rejected pixels... for a trail like at satellite..it makes a fatter trail.

The rejection maps that you get include this final rejection with the growth. If you want to see the difference. Run Image Integration without Large Scale Rejection and you will have the rejected pixels that come only from the statistical analysis. You can blink the images...and be clever and take the difference of the two rejections maps.

Let me also add that I take advantage of this very behavior for my Selective Rejection technique. For example, you would not need to calculate a line for the satellite or anything else. You just need to paint black pixels (with the clone tool) over anything you want to reject. It only takes a few seconds to do.

-adam

[pfile]

[edit] aw, forget it.

[ngc1535]

I am sorry Rob. edit>*forgotten* Your answers are so good they often need to be unpacked! (I think).
It seemed to me the original poster and your answers were veering.

So you are saying that a satellite trail (when it is in the form of the rejected pixel map that is analyzed using multiscale algorithms) might not be large enough to be grown?

I think comparing the resulting maps (before and after Large Scale Rejection) most definitely answers the question. It shows if the satellite is being captured... if increasing the number of layers helps..etc etc. I have found that satellite trails of 2-3 pixels in size are detected and the growth occurs. I have not tested on satellite trails of 1 pixel dimensions. For the northern hemisphere M42 is a particularly busy area with all of the geostationary satellites running through the field. 

I have never gotten large scale rejection to take care of dust donuts because they are so rarely ever completely rejected initially.  Really big donuts can be somewhat helped with Local Normalization ... but they need to be quite large and only part of a set of images so there is a good reference frame generated.

adam

[pfile]

i am saying that i've tried to reject satellite trails with LS rejection and never had complete rejection despite tweaking the trails... the only thing that's worked for me is clobbering the data as you suggest.

in day to day usage for me i end up having 100s of subs and so any residual unrejected pixels just fade into the average, so in practice it's not a problem. but for some dark-sky work where the sub SNR is high and the number of subs is low, i frequently had this problem. so i would just do the pixelmath thing since the # of subs was not onerous.

rob

[gregwjones]

While I do appreciate the discussion about removing sat trails using large-scale pixel reduction (LSPR), my primary question is: 

Should I get a separate LSPR image (rejection map) from the ImageIntegration process?

I currently do not see one.

Regards,
Greg

[pfile]

no i think adam explained that the results of the LS rejection are part of the rejection map you get. to see the deltas you'd have to run with LS rejection turned on and turned off and compare the two rejection maps, perhaps by subtracting one from the other with pixelmath, or perhaps dividing them.

rob