"non-traditional" use of GradientsMergeMosaic

[pfile]

so yesterday i decided to try to merge in the M42/NGC1977 area from a 200mm widefield image into a 50mm widefield image of Orion. the M42 in the 50mm widefield was pretty blown out and colorless.

i cropped out just the M42 area and registered it to the 50mm image. i then struggled with pixelmath for a while, trying to get the blend right. but then it hit me - why not try GradientsMergeMosaic?

so i did that, and the result looks great (though a bit funny due to the higher resolution of m42 + the original had undergone MT and so the star profiles don't match). still, zoomed out it looks much better than the original.

however, the resultant image is a bit desaturated - the bright stars from the original image have lost their color. perhaps this is a result of the black background surrounding the crop of M42.

attached is a closeup of the blend. looks great aside from the FHWM/resolution mismatch.

http://www.flickr.com/photos/pfile/8377462056/

does it make sense to fill the cropped image with, say, the median value from the 50mm image instead? or is this just unavoidable due to the algorithm involved?

[georg.viehoever]

Hi,

I am using  GMM as a debloomer in http://pixinsight.com/forum/index.php?topic=5023.msg34770#msg34770, basically pasting neutral data into bloomed areas. What you are apparently doing is pasting an actual image into overexposed areas, something that is perfectly legal - no "painting" involvled. GradientHdrComposition would also be an option, but I am not sure how well it works on non-linear data. In general, GMM can be used to inpaint anything into something else. I guess some of the selfhealing brushes of other programs work similarly.

I also noticed that sometimes the result of GMM seems to loose contrast/saturation. In my opionion, this is because GMM usually removes extreme values at either end of the scale (0=black background: gone after GMM, bright stars: often go away because of feathering). One of the last commands of the GMM implementation is Rescale(), which basically restretches the image to the range [0,1]. This usually causes bright parts of the image to be in a flat region of the STF, causing low contrast and low saturation. In reality, the  data is still as linear as it was before. That's my theory, but I did not thoroughly investigate it.

Maybe this bit of the GMM needs a re-design to preserve the original appearance as much as possible.

Georg

[pfile]

thanks for the reply georg. so i guess then the pixel values of the "non-active" area of the M42 image do not matter then?

i wonder if it makes sense to mask off the rest of the image while doing the GMM. i'll have to do some experiments.

rob

[georg.viehoever]

Rob,
feel free to do experiments, and report your results. I have the intention to do a GMM2.0 later this year, based on what we learned so far.
Georg