Getting different results on preview versus whole image

[astrovienna]

When I apply the same settings to the whole image, I should get exactly the same result, right?  I'm following Alejandro's approach to noise reduction described here:

http://pixinsight.com.ar/en/info/processing-examples/13/m-97-owl-nebula.html

using GreyCStoration and ATWT (actually I compared ATWT to MLT using identical k-sigma thresholding settings, and slightly prefer MLT).  It gives a beautiful result on a preview of a galaxy I'm working on.  But when I apply it to the whole image, it's much, much worse.  I get the dreaded lumpy oatmeal effect, and the noise reduction isn't anywhere near as effective.  Here is the (cloned) preview and a closeup of the same area after the processing was applied to the whole image.

http://www.pbase.com/skybox/image/160423918

Any idea what I could be doing wrong?

Sorry for the flurry of questions recently.  I'm working through my first LRGB with PI, and each little step forward seems to create new issues.  The narrowband image I did was simple compared to this.

Kevin

[oldwexi]

Hi Kevin!
No, you cannot expect exactly the same result of the preview in the whole mage.

I recommend to create a mask which covers the bright parts and let the MLT only work on the darker parts of the Image.

I am just building an Austrian PI-Tutorial Page in german language.
In the moment it is in design test stage, however your issue is already covered here:
http://www.alrukaba.at/cms/upload/PI/Entrauschen_mit_MLT_Main.html

In the example you can see that the preview is placed over the darker parts also.
The brighter parts usually arent noisy anyhow.

Alles claro?
Gerald

[astrovienna]

Hi Gerald,

I understand that the effect on the preview area might not be representative of the effect on other parts of the image, but for the previewed area itself, shouldn't I see the same effect when I apply to the whole image?  Because if not, it kind of defeats the point of applying a process to a preview, at least when the difference is dramatic, as it is here.  Does the image I posted

http://www.pbase.com/skybox/image/160423918/original

look like what you would expect?  I haven't processed a lot of images in PI yet, but I've never seen anything like this.

Kevin

[msmythers]

Kevin

I think this posting will help you to understand a little the issue you are seeing. http://pixinsight.com/forum/index.php?topic=8461.msg55085#msg55085 . Remember the preview is something you define not the tool. If a tool works on one pixel at a time say like Curves then the preview will be correct not matter what size or where the preview is in the image. If the tool works on pixels based on scaling or other pixels within the image the preview might not look the same as when the tool is applied to the entire image.

There are many postings within the forum on this subject. The basic question comes up with new users who are not familiar with this type of processing. This is one of the big differences between PI and painting programs. The quality image results I think speak for themselves. It just takes a little time to learn. 

So knowing the way the tool functions is what you will learn over time. Also how to achieve your intended results with the use of masks while working with certain tools is a must.


Mike

[astrovienna]

Mike, thanks very much.  I think I get it now:  multiscale algorithms are scale dependent, and previews may not include representative scale structures.  I will note, however, that Juan specifically said that while that applies to HDRMT, it does not apply to MLT or MMT, because PI has built-in some special routines in those tools to get around that issue.  And it's MLT that I'm using here.

More important, I've now re-created the problem on a full scale image by creating a "preview" that includes the entire image.  When I apply the MLT to that preview, I get the desired effect (and wow, it is REALLY impressive!).  But when I apply it  to the image itself, I get the much lower quality result shown in the side-by-side comparison image I posted earlier.  For the life of me, I can't figure out why that would be.   The effect on the full image really should match the effect on a full-scale preview, shouldn't it?  I'm sure I must be doing something wrong.  However, I can work around it by cloning the "preview" and doing all further work on that clone.  BTW, I have been using masks all through the processing of this image.

If anyone would like to try to reproduce this result, I'd really appreciate it.  The image I'm working on is linked below, at the point right before I apply GreyC and MLT:

https://dl.dropboxusercontent.com/u/58468743/LumwithMay81_DBE_DBE1_forcropping.fit

I'm afraid I'm more mystified than ever, but at least I have a work-around.

Kevin

[msmythers]

Kevin,

Are you using the linear mask in MLT and setting the parameters for the mask with SFT applied while looking at the whole image? If so try setting the mask without STF. Then you can reapply STF after setting the mask parameters. Just a thought.


Mike

[astrovienna]

Hi Mike,

Thanks for the quick reply!  I'm not using the linear mask in MLT.  I found that making my own mask was giving better control over the result.  I do have STF on whether I'm applying MLT to the preview or the image itself.

Edit:  So I just realized that means I'm applying a non-linear mask to a linear image before MLT.  I don't know how that would give this result, but a lot of this is just a black box to me.

Kevin

[msmythers]

Kevin,

Try this MLT icon and see if it doesn't give the same results for both the main view and the preview. You can double click on the icon to see the settings. I set the linear mask with STF off. Then after turning the mask preview off I applied STF. I then made my preview and made my MLT settings. 

Mike

[msmythers]

Kevin,

Don't know if you've seen this example of MLT and the linear mask? http://www.pixinsight.com/examples/M81M82/index.html#Noise_Reduction It's my go to when someone needs to know how to use MLT for noise reduction. I figure it from Juan so it should be right, haha.


Mike

[astrovienna]

Yes!  That's exactly the page that started me down the path of using MLT for noise reduction instead of ATWT or TGVD.  However, I just recently came across this processing tutorial by Alejandro:

http://pixinsight.com.ar/en/info/processing-examples/13/m-97-owl-nebula.html

which led me to use the k-sigma noise thresholding panel in MLT instead of the noise reduction panel that Juan used.  I suspect that if I started with Harry's excellent data, like Juan did, I wouldn't need to do that because background lumpiness wouldn't be an issue.  But I shoot through Bortle 8 skies.  I have no idea how the two procedures differ in theory, but the difference in results borders on the miraculous.

Kevin

[msmythers]

Kevin

I have better results for most of my images with MLT without K-Sigma but that's just me. I shot with poor equipment from a Bortle 7-8 area also. On top of the light issues I have very poor seeing most nights of the year as I'm on the east coast of Florida with very heavy humidity. I've just come to expect that I will have noise  because of conditions more then software issues. I'm just happy to be able to image. For me it beats trying to look through an eyepiece. You can see some of my results on my Astrobin page.

 

Mike
http://www.astrobin.com/users/msmythers/

[astrovienna]

Try this MLT icon and see if it doesn't give the same results for both the main view and the preview. You can double click on the icon to see the settings. I set the linear mask with STF off. Then after turning the mask preview off I applied STF. I then made my preview and made my MLT settings. 

I tried your MLT icon, and as near as I can tell the preview and main image look the same.  However, that's very different from the MLT process I've been using.  Try the two icons (GreyC first, then MLT) attached here on the image I uploaded, and see if you don't get very different results between preview and main image, even when the "preview" is the entire image.  BTW, I tested both with the MLT linear mask option and with my own (non-linear) mask, and the issue shows up with either approach to masking.

https://dl.dropboxusercontent.com/u/58468743/LumwithMay81_DBE_DBE1_forcropping.fit

Kevin

[msmythers]

Kevin

Are you applying GreyC and then MLT to the same preview without applying to the main image? What exact order are you applying the tools? If I apply GreyC to the preview first and then to the main image they look the same. Then I can apply MLT to the preview and then the image and they look the same. If you want to see what both GreyC and MLT with look like in that order before applying to the main image you have to use the "store" command for previews after you apply GreyC. Otherwise the next tool you apply to the preview will be the only tool applied to the preview.

So what I am seeing if I apply GreyC and then MLT to the preview without the store command is a very smooth image, only MLT is applied. If I go and apply GreyC and them MLT to the main image I do not see that smooth image. I would expect that. If I apply only MLT to the main image I see the same smooth image I had in the preview.



Mike

[astrovienna]

Yes, that is exactly what I was doing wrong!  Mike, thanks so much for figuring this out.  I was ready to insert "clone preview of whole image" into my processing routine.  I was using both GreyC and MLT because Alejandro used both GreyC and ATWT (which looks very similar) in his M97 tutorial.  Now to figure out what settings I really should use on MLT.  I can see that I have at least one image that I'm going to have to go back and reprocess now that I understand this a bit better.

Kevin