Integration not overlaying images

[jcinpv]

I have a set of 12 Lights of M3, 10 Darks, 100 Bias, and 28 Flats. All are at the same ISO (800).
Since the images are all taken with a Canon 60Da, I processed all images with CanonBandingReduction before working with BPP.
Individually, the images look fantastic!
Then I ran ImageIntegration on the 12 registered and calibrated images.
Ugh!
It appears that the images did not get matched up.
I've tried both Multiplicative and Additive with scaling with Noise evaluation. The scale estimator is set to Iterative k-sigma/biweight midvariance.
I'm using Windsorized sigma clipping with Scale + zero offset.
What have I done wrong?
What information do you need in order to help me figure this out?
Image of Integration is attached.

[georg.viehoever]

Check if the images are properly registered. You can use the Blink process for that.

[jcinpv]

I guess you mean that if they are properly registered then Blink will show all images exactly one on top of the other. They aren't.
So could I have done wrong with the BPP script?

[jcinpv]

Here's a snapshot of the Registration parameters in BPP:

[Carlos Milovic]

Try to run StarAligment on your calibrated subs. Then integrate the result.

[georg.viehoever]

StarAlignment also has a number of modes that help you to tune its settings (Detected Stars, Matched Stars mode). My guess is that your star detection settings see a lot of non-stars (for instance hot pixels, noise), and then somehow match those noisy pixels instead of the stars. You may need to tune Sensitivity and or Noise scales. See http://pixinsight.com/doc/tools/StarAlignment/StarAlignment.html#usage_002 for details.
Georg

[jcinpv]

Thanks, guys. Star Alignment followed by Registration seems to work. Got to mess around with parameters in SA to get all 13 images. First pass got only 4.

[Harry page]

Hi

I have found with this sort of data there is a need to increase the noise scale to avoid false star matches ( star alighn on hot pixels etc)

don't be afraid to say try a fig of 8 to 10


Harry

[jcinpv]

I've tried several possibilities of several different tweaks. If the RANSAC tolerance is set beyond the default of 2, the images start to warp (stars in the center stay pretty much put, but the stars around the edge wind up in swirling about; it's like the images get rotated from one to the next). It if is set below 2, only one image is the result. The other settings of Detection Scales and Noise Scales makes no difference to the result. The best I get is 5 solid images out of 13. And there are still hot pixels left in the final registered image.

[Carlos Milovic]

Then, you may try DynamicAligment. Set the star couples for the first reference/target, and then create a process icon. You may use it later to load the set of couples to another target, using the same reference.

[jcinpv]

Then, you may try DynamicAligment. Set the star couples for the first reference/target, and then create a process icon. You may use it later to load the set of couples to another target, using the same reference.

Thanks, but I haven't gotten to that lesson yet, and what you just said went over my head. "Star couples?" I opened DynamicAlignment and couldn't even figure out step 1. So which PI lesson is that in?

[Carlos Milovic]

DynamicAligment is the manual way to align images. First you set the reference and target frames (by clicking on them). Then, click on a star in the reference frame. A sample will also be created on the target frame, in the same location. If your frames are close enough, the automatic star detection will centre the samples. If not, you'll have to manually place it on the same star, in the target frame. Then, add another sample on the reference frame and so on. Each time you add a new sample, the algorithm learns from the previous ones, so it falls closer to the desired star on the target frame.
To avoid creating all the samples, when working with several frames, you may create a process icon at the end, to reuse the set with another target. You'll have to open DA again, select the same reference and the new target. Then drag the process icon and drop it on the bottom bar in DA's window.

Hope this helps.

[jcinpv]

DynamicAligment is the manual way to align images. First you set the reference and target frames (by clicking on them). Then, click on a star in the reference frame. A sample will also be created on the target frame, in the same location. If your frames are close enough, the automatic star detection will centre the samples. If not, you'll have to manually place it on the same star, in the target frame. Then, add another sample on the reference frame and so on. Each time you add a new sample, the algorithm learns from the previous ones, so it falls closer to the desired star on the target frame.
To avoid creating all the samples, when working with several frames, you may create a process icon at the end, to reuse the set with another target. You'll have to open DA again, select the same reference and the new target. Then drag the process icon and drop it on the bottom bar in DA's window.

Hope this helps.

Well, once I deduced that after opening DA, I needed to open a calibrated image to use as a reference frame, then open another calibrated image to use as the target frame, then start selecting stars in the reference frame and watch the same stars get selected in the target frame. What's fun about following someone's directions is in figuring out the missing detail.

I managed to get 11 out of the 13 to stack. The result is bit deformed as compared to the original reference. And now I've got these strange diagonal and vertical gradients all over the image even after DBE and BN was applied. Well, I've finally moved away from Integration, so maybe I'll continue this in another thread at another time.

[Harry page]

Hi

Can you share these problem files with me as I would like to see what the problem is

Harry

[jcinpv]

Harry, you can access any or all of the files here: https://copy.com/B44Scd6EGfoQ4ucP
The set includes all the original Lights, the Masters (Bias, Flats, Darks), the Calibrated, and the Registered.
There's a single shot of the originals that I manipulated a bit with Preview on my Mac.
14.24 GB.

Take what you need. (This is public, so if anyone else wants a crack at it, be my guest.)

John C.