HDRMT on very large images

[robhawleyastro]

Many writers have described Previews as not producing the same result as on the full image.  My case is almost the opposite.  The preview is producing the results I expect and the images are being destroyed on the full image.

First this is a huge mosaic (31127 x 16982) the image contains a significant border due the the distortions of a far north mosaic.
http://robhawley.net/pixihdrtransform/overview.png

If I create a preview of the M52 / Bubble Nebula region I see what I expect

original
http://robhawley.net/pixihdrtransform/preview_original.png

median transform
http://robhawley.net/pixihdrtransform/preview_median.png

B spline
http://robhawley.net/pixihdrtransform/previewBspline.png

However if I do the same operation on the full image the Bubble is washed out instead of being enhanced. Cropping to just area of the preview following the HDRMT I get
http://robhawley.net/pixihdrtransform/full_median.png
http://robhawley.net/pixihdrtransform/full_Bspline.png


On the theory that either the black border or the size were the culprit(s) I cropped to eliminate the border and reduce the size.  Just taking the center section ~(18000x7000) worked better . B spline now works, but median makes little change to the image.  Cropping further to 8000x6285 allowed both to work as shown in the preview

If someone wants to play with the full image they are here.  Warning even compressed these are Gbyte files so the path is to the folder instead of the image.
http://robhawley.net/pixihdrtransform/

If this is pilot error then I would like to know what I am doing wrong.  I tried this with and without masks (I included some masks in the folder).  Other than that I am using typical settings; to lightness and lightness mask.  If this is a limitation in HDRMT then suggestions of how to work around it.

[vicent_peris]

Hi Rob,

The problem occurs because you have black areas since you didn't finish the mosaic. The mosaic panels have very sharp edges that affect the rescaling of the wavelet layers. Once you finish the mosaic and crop the image to avoid the black edges everything will go well.


Best regards,
Vicent.

[msmythers]

Vicent

Just a question, if you created a mask of the black border region and used that to block the border would that work?



Mike

[robhawleyastro]

Vicent  that is what I first suspected also.  Why then did I have a problem with a central section that lacked the black border?  Reducing its size made it work correctly.

Could a do a workaround where I set the black area to a background level shade then then run HDRMT?

msmythers - no  my masks already masked the boundaries so they would not be affected.  Vicent is discussing what I originally suspected as the problem.  The sharp edges and the black border are disrupting the algorithm.

[vicent_peris]

Hi,

The problem is that these mosaic edges are absolutely sharp, thus they have an "infinite" contrast from a multiscale point of view, unbalancing the dynamic range compression. There is no solution for that right now; I have a possible one, tough, that could be implemented in the future. For the moment, just crop the mosaic once it's finished.


Best regards,
Vicent.

[robhawleyastro]

OK I have at least a partial workaround.

By using clone stamp I removed the sharp edges.  The image now appears to fill the entire frame.  With this change B Spline works fine.  The median transform still does nothing, but as I mentioned above, I saw the same time when I worked on an image that cropped the black, but was still large.

This is an acceptable outcome.  I can restore the true limits of the image before I finish it.

[robhawleyastro]

I cropped the image again this time rotating the crop so I could get a wider image.  The new one is 18245 x 5848.  This is wider, but narrower than the previous test.  Median worked fine on this image.  Apparently I must have left an edge in the previous crop test.