Differences between ATrousWavelets Transform versus MultiscaleMedian Transform

[jerryyyyy]

Hello,

I have been dutifully following the information in Jordi's Powerpoint on noise reduction and there he used BOTH these procedures.  I cannot find any documentation on the later and the interface looks the same. 

What gives here?  Is one a later version of the other?

Thanks

PS  Also in this series, whcih I generally find very helpful, he uses the Extract Wavelets Layer Script to somehow figure out the settings for ATW... these seems sorcery to me.... no idea how it happens... though it must be a good idea. 

[Carlos Milovic]

Hi Jerry

Both are multiscale tools, that decompose the image into layers that contains information at different scales. That means, objects of different sizes are "stored" in different layers. The difference between them is how the separation between the scales is performed. Basically, ATWT uses convolutions (bluring), while MMT uses the median operator (which is a non linear operator). In practice, MMT is better at detecting objects with sharp edges, and small scale objects, while ATWT is better at larger scales, or smoother objects.

I recommend you to take a look at a comparison Juan made about different noise reduction tools, and the new TGVDenoise, in the release information subforum.

[jerryyyyy]

Hi Jerry

Both are multiscale tools, that decompose the image into layers that contains information at different scales. That means, objects of different sizes are "stored" in different layers. The difference between them is how the separation between the scales is performed. Basically, ATWT uses convolutions (bluring), while MMT uses the median operator (which is a non linear operator). In practice, MMT is better at detecting objects with sharp edges, and small scale objects, while ATWT is better at larger scales, or smoother objects.

I recommend you to take a look at a comparison Juan made about different noise reduction tools, and the new TGVDenoise, in the release information subforum.

Thanks very much.  I am working on a project to try to visualize the external "coronas" of M57 from my very noise polluted site using 1h H-alpha images... now up to 18h.  I worked it over with TGVDenoise following the instructions, and you are right it is superior.  I guess I have to try it out setting by setting at a high magnification to see the actual results.  I know I have a signal out on the outer corona since I can check pixel by pixel.  Any suggestions on optimizing TGVDenoise appreciated.  What I am doing is to run it on the delinearized image about 2-3 times with stretching between.  It is not clear whether using any other procedures helps or hinders... I seem to think more likely to hinder and introduce noise.