Atrous & MMT Extreme Frustrations

[pscammp]

Hello all,
     Why is it that the ScreenTransferFunction does a beautiful job of displaying
my linear data until I use either ATWT or MMT, after this, if I reset the STF and
do an Auto stretch again the result looks like the contrast has been cranked up
through the ceiling.

In case anyone wonders:
I always reset and re-apply the STF after every step in the Linear processing stage,
the colour calibration stage is a typical example, after this is applied, the whole image
goes bright blue, on re-linking the channels and re-applying the SFT all looks the way
it should and wonderfully colour balanced (for want of a better expression).

What am I doing wrong that a little Noise reduction in the Linear stage using ATWT or
MMT completely screw's up the STF display 

Image Before ATWT:

https://dl.dropboxusercontent.com/u/4938813/Before_ATWT.jpg

Image After ATWT:

https://dl.dropboxusercontent.com/u/4938813/After_ATWT.jpg


Please tell me it's something im doing wrong 

Cheers
Paul



 

[Geoff]

The STF does mothing to change the data. It simply displays the data you have, so it has to be ATWT or MMT that's screwing up the data.
Geoff

[pfile]

i guess the thing to keep in mind here is that STF does a statistical analysis on the image to determine how much of a screen stretch to apply.

after noise reduction the statistics of the image are different and so STF ends up computing something different. i guess if you want to see what the image looks like with the old STF you can save the STF before the noise reduction and then restore it afterward.

in fact if you save the new STF and undo the NR, you'll probably see that the data looked like that all along, but STF was hiding it from you.

rob

[pscammp]

Cheers Guy's
   
Geoff,
    I agree

Pfile,
    Ok so this is what I have done on your suggestion:

See this image file:
https://dl.dropboxusercontent.com/u/4938813/ATWT_Issue.jpg

1. Take x2 clones of original image for later

2. Opened my previously stacked image

3. Apply autostretch with STF (channels unlinked due to a red bias in the image)

4. Saved a STF process icon to desktop called 'BeforeATWT'

5. Cropped the rough from around the edges

6. Apply ATWT noise reduction to layers 1,2,3 and 4 with Thresholds set to 3.0, 2.0, 1.0 and 0.5 respectively

7. Reset STF and Autostretch again which gives the weird result

8. Create a new process icon called 'AfterATWT' for the STF's new state/calculations

9. Drop the 'BeforeATWT' icon onto first clone image (upper right image in my attached image)

10. Drop the 'AfterATWT' icon onto second clone image (lower right image in my attached image)

Ref my linked image above:
 So the big image on the left is the image after ATWT reset and autostretched again, not sure if you
can see it but there is quite a bit of gradient lines in the background image, whether this should be
expected or not is one for the experts but hitting the '24' icon sorts this issue out visually.

To the images on the right:

Top one is as it was autostretched with the original 'BeforeATWT' icon so this is where the image looks
before ATWT was applied.

The bottom one on the other hand shows a difference to the 'Original' stretch obviously due to the subtle
differences in the autoSTF settings because of the change in the pixels from the before and after ATWT.

BUT, there is one thing I do find strange..... Looking at M33 itself, it looks like colourCorrection has been
applied as it has changed to a bright white. Also M33's brightness compared to the original has intensified
a hell of a lot. Maybe this is what is making the STF produce this aggressive contrast impression im pointing
out in this thread. At the end of the day, all ATWT has done is kind of blur pixels into each other in essence
to remove the appearance of noise, this will obviously cause a change in the image which the SFT has to
analyse but the result looks so extreme.

Cheers
Paul

   
 

[pfile]

well with respect to the color balance, if the channels are unlinked in STF then the image will tend to look "correct" as the unlinked channels has the effect of balancing the histograms. again due to the different statistical properties after the noise reduction it's reasonable that the balance would look different.

anyway i am not sure it's too much to worry about. probably on this data the first thing to do is a very careful DBE to try to get rid of the weird gradients in the background, then do background neutralization and colorcorrection, then do the NR and the rest of the processing.

rob