Stretching vs Contrast Enhancement Processes - GHS and Stretching Academy

David Payne

Hi Adam,

Great series on stretching and introduction to GHS.  I think your Stretch Academy Series is just great.

One point that I think it is necessary to clarify is a point you made on choosing the path you take.  In the latest video (April 26th), I believe you were comparing "your" typical path, versus the "GHS" path, but I think you may have been actually comparing a path that would include LHE or HDMRT with one that used GHS alone.

To me, these are not a valid comparison between GHS and other stretching methods, such as HT, masked stretching, or curves - but a rather, a comparison between using and without using LHE/HDMRT processes to "artificially" enhance contrast.   I consider LHE and HDMRT very important and useful processes, but I don't consider them "stretching".   Stretching takes the limited contrast available to us and applies it in a manner that does not compromise the data, ie, makes the best use of the available contrast.    

LHE, and HDMRT employ a mind/eye trick that reduces contrast from one scale (for HDMRT) or one area of the image (LHE) so that it can be applied it to another.   It then appears, based on the way the mind and eye work, that you have actually enhanced the contrast.   Using LHE and HDMRT, the rank order of pixel brightness is destroyed in the image (there is no going back) while mere stretching can be performed without changing this rank ordering (what is brightest in the image will remain the brightest in the output).

For visual consumption, I almost always use some degree of LHE and/or HDMRT in my image processing.   I use GHS both before and after LHE/HDMRT with a close eye on the histogram.   I can replace other stretching processes (masked stretch, HT, curves, arcsinh) completely with GHS, but I cannot replace LHE/HDMRT because it does something fundamentally different.   Whether you use LHE/HDMRT does not depend upon whether or not you use GHS or not.   In this way, I think your comparison in the last video on the Great Orion Nebula was a little bit misleading.

Otherwise, though, fantastic series and congratulations.

Dave

BlockHead

Dave,

We can have a more detailed discussion off line. I do not think I was being misleading.

My point was that a path of HT MTF followed by HDRMT may be shorter/different than attempting to manage the contrast at the high end via GHS which is something that you in particular point out as one of its strengths.

Throughout the series I am striving to indicate different ways to achieve results. If, in the case of some large dynamic range targets, all that is necessary is to get that initial stretch in faint part of the histogram- the differences between the HT MTF curve and another hyperbolic curve with a large local intensity inflexion are not terribly different in the initial stretch. GHS really has power to shift contrast around at more well-defined brightness levels when the histogram (having been stretched) fills more of the brightness-frequency space. Using GHS to remove too much contrast from the top end (as happens by modifying the local intensity) is counterproductive for HDRMT. You can achieve a similar result that HT MTF stretching gives you... but potentially in more steps at the risk of being counterproductive.

The future use of HDRMT is driving the result of what the input to it needs to be like. And this path to this kind of input image is different in length and adjustment. This is what I was pointing out. It *does* depend in my opinion on whether you use GHS for these reasons. Again, you can certainly achieve the right kind of input image for HDRMT using GHS- but the path is different. 

-the Blockhead