MMT versus MLT for enhancing details?

[Buzz]

I know that experimentation is always the key and I do as much as time allows. I'm going deeper into PI with excellent results but sometimes I need a little help to confine the experimentation.

For the subject of sharpening - I notice that many tutorials discuss MLT or MMT exclusively but never directly compare.

I know there is always a sound numerical basis for PI tools. Are there any suggestions on which one to use and at what point in the workflow?

[Alejandro Tombolini]

Hi Chris, this post and other reference there have some information that may help.

Saludos, Alejandro.

[Buzz]

Sorry, I'm missing something - these posts are on noise reduction. I was more interested in the sharpening processes. For instance:

The following refers to MMT  http://www.pixinsight.com/tutorials/mmt-sharpening/index.html

and this one uses MLT, after deconvolution.

http://www.lightvortexastronomy.com/tutorial-sharpening-fine-details.html

I was wondering if MMT had overtaken MLT or there were specific instances to use one over the other.

[tsaban]

Hello Chris,

I have recently used MLT for sharpening the corona of my solar eclipse images. I did use a mask to protect low S/N areas and then carefully increased the bias slider in the 'Detail Layer' area up to layer 4. The corona was nicely sharpened. Of course one has to be carefull not to boost noise, so the application will depend on the image data.

Clear skies
Tahir

[msmythers]

Buzz

As a note MMT was released in 2011 and MLT in 2014. MMT was the replacement for ATrousWaveletTransform. MMT has had updates since 2011. I am not sure if MLT has fully replaced MMT in all functionality at this time. Juan would be the one to answer that question. MLT has added features that are not available in MMT. Originally MMT did not work correctly for medium or large scale sharpening. I don't know if that was corrected or even possible in MMT even with later updates. https://pixinsight.com/forum/index.php?topic=3552.msg24497#msg24497

One thing to keep in mind with all tutorials, look at the date they were released if possible. A single tutorial is rarely updated to use new tools. There are lots of now old tutorials out there. You will even find some newer ones using outdated tools like ATWT or ACDNR. That's fine because that is what the author is comfortable using and they are showing how they processed a particular image. Old tools and some scripts are kept in PI for archival purpose meaning if someone has a project from 2013 they can open it and continue to work on it with the same tools. Also if your comparing data capture today to data captured and processed 4 years ago you might need to use the same tools to be able to compare results properly.


Regards

Mike

[Buzz]

Juan  - can you comment on the applicability of MMT and MLT for sharpening and where the older MMT is preferred over MLT for this purpose? It would be really interesting to hear your views.
many thanks

[jkmorse]

Chris,

See Juan's comments in this post, though it applies to noise reduction and not sharpening.  But his discussion may provide some clues re what you are after:

http://pixinsight.com/forum/index.php?topic=8105.msg53541#msg53541

My take away is that both remain viable tools (in the last line he says the same about sharpening), but do their magic in different ways.

For what it's worth,

Jim

[Juan Conejero]

Juan  - can you comment on the applicability of MMT and MLT for sharpening and where the older MMT is preferred over MLT for this purpose?

As Jim says, both are viable tools. However, there some important differences and facts that should always be taken into account:

- MMT is a ringing-free transformation. See this tutorial. This property makes it particularly suitable for local contrast enhancement (aka sharpening). MLT, as any linear transform, suffers from ringing, although our deringing algorithms alleviate this problem.

- The counterpart to ringing is small artifacts that can be generated by MMT around image structures that are very different morphologically from the structuring elements used to apply median filters. With our implementation, however, this is usually a marginal problem, and even more marginal with astronomical images, where sharp linear edges and angles are rare. The key word here is careful application.

- MMT works very well to sharpen small-scale structures. However, it does not work to enhance structures at medium and large scales, where it can generate large (and often not obvious) artifacts.

- The linear transforms implemented in MLT, including the Starlet (aka á trous wavelet) and multiscale linear transforms, have the important advantage that they are isotropic, that is, they modify the image exactly the same way in all directions. This is not true for MMT, although our implementation is reasonably close to isotropic. Isotropy is a very important property to process astronomical images.

- MLT works very well at medium and large scales. It does not work so well at small scales, mainly as a result of its poor efficiency to isolate structures, compared to MMT.

- The Wavelet-median transform, which I implemented in MMT a few years ago, attempts to provide the best of both linear and nonlinear worlds: MMT for small high-contrast structures, and MLT for large-scale, low-contrast structures (such as the background).

Experimentation is always necessary, as well as knowing that each image poses different problems requiring different solutions. I hope this helps.