The thing to remember is that neither Landsat-8 and Sentinel-1 is single purpose. Polar scenes by their nature have very poor intrinsic contrast (off-white on white). Further, some users need the actual instrumental luminance values (eg band 10 Landsat thermal for ground temperature). Together this means the products provided are not optimally adjusted as downloaded -- a jungle scene has different issues from an ice stream.
The first round of contrast enhancement is usually histogram normalization. That is, the image is not making full use of dynamic range (eg 20-190 instead of 0-255 on an 8-bit). That's easily fixed by adding a constant and multiplicative rescaling (brightness and contrast knobs).
That is often followed by a second order adjustment of mid-tones (gamma correction). This can greatly improve the image visually but in general loses the original 'scientific' content. Contrarily, functions like log or sq root can have a logical rationale but they seldom produce a favorable visual outcome. (Gimp has a plugin that allows *any* function to be implemented as its power series expansion.)
The next level of contrast enhancement is histogram equalization. This uses the cumulative distribution function (see wikipedia) of pixel intensities to flatten the histogram (ie expand contrast for well-represented pixel values). This is quick and very effective on both Landsat and Sentinel but using the global histogram again means discarding neighborhood information.
Thus neither method is optimal for detecting a developing fracture (Espen's goal in #645) since neighboring pixel correlation is important to that.
It's better here to start with adaptive histogram equalization (CLAHE, see wikipedia). That is provided in Fiji/ImageJ2 at the bottom of the 'Process' menu; see
http://fiji.sc/wiki/index.php/Enhance_Local_Contrast_%28CLAHE%29 The image below shows its effect on Landsat's low-budget preview image that Espen likes because of its small file size.
It is imperative to work within the 16-bit world of Landsat and Sentinel for initial contrast adjustments and co-registration rotations. Otherwise you will get unoccupied bars in the histogram.That cannot be currently done within Gimp. However it is easy to do a quick in-and-out within ImageJ and return to your usual image processor with an 8-bit after the benefits of 16-bit have been exhausted..
For a linear feature with predictable orientation (~parallel to the calving front), an edge detector would be the natural follow-on. These are typically directional derivatives or convolutions.
However nothing you can do will make a silk purse out of a sow's ear (Sentinel EW). Ultimately there's no substitute for a really high resolution image, one commensurate with the task at hand.
I'll post a 0.5 m resolution DigiGlobe image of growing fracture tips over at the Petermann forum. Even prior to polar-appropriate contrast enhancement, it's a huge improvement over the 15 m band 8 of Landsat, even if you've applied CLAHE or retinex in Wipneus's pan-sharpening protocol.