[irwjager (Ivo)]

Originally Posted by rcheshire

Ivo. It is a case of protection. Wavelets are quite destructive without a mask, although the new Median wavelet transform tool is much better and pretty much ring free - still I would use a mask keeping other structures intact.

Understood. I was just wondering if there was any reason you choose smooth luminance mask (causing darker/brighter areas to respond differently) over an on/off type mask.

If you have a look at PI, you will notice that most of the native tools are wavelet based, and very effective for noise reduction and sharpening linear and non-linear images.

Hehe, yeah I can't help but think they've gone a bit wavelet crazy I mean that in a 'to a man with a hammer every problem looks like a nail' sort of way.

Not sure how that applies to ST, but wavelets are wavelets.

To be honest, Wavelets as found in PI and Registax are one of those tools that have always bothered me. They can be immensely unpredictable if you don't know what you're doing (and let's be honest most people have trouble grasping the math behind them and I can't blame them!). The end result is often an unnatural looking image with a dynamic range swinging all over the place; appropriate in some places in the image, destroying contrast in others.

This is because, by default, Wavelets have no regard for detail in an image They don't know where, in an image, to ease off or where to ramp it up. Bad luck if the flute and violin occupy the same frequency band - they all get the same treatment. The flute may sound great now but the violin's subtle overtones are drowned out by the comparatively louder prime note. In image terms the large spiral arms stand out more now, but the finer details within them have become harder to see (contrast has been reduced).

The bottomline is, "scale" alone is a poor way of isolating features in an image (random noise excepted!). If you have other, more intelligent means of identifying and isolating the features that you're after, then use those (smarter algorithms or user generated masks). If that is not a possibility or you can live with sub-optimal results, then sure, use wavelets.

Rowland, your example of using a mask and only then using wavelets is a very appropriate way of reining them in somewhat.

You can tell I'm passionate about this and I've indeed been quite reluctant to add 'PI-style' wavelets to ST for aforementioned reasons; they are very user-unfriendly, unpredictable and usually are a poor tool to solve your problem (bringing out detail).

ST uses a more up-to-date set of algorithms for multi-scale aware dynamic range optimization (the Optimize module) where it actually takes into account whether an individual pixel basis stands to gain from brightening or darkening across an infinite number of scales/bands. It determines the best-case (consensus) brightness level for a pixel based on all possible scenarios rather than picking just a single scale. The result is a more natural looking image where dynamic range is optimised across all scales. Best of all, the computer does all the hard work and the user doesn't have to tinker around with masks and abstract/nondescript sliders.

In musical terms, it automatically adjust the equalizer whenever the violin is playing to capture every nuance of the instrument, quickly changing to another optimal setting for the flute. Should both be playing at the same time, it picks the 'best compromise' setting that still enhance both without ruining the performance of either.

Cheers,