Histogram Shape and Contrast

Craig Kolb

At the 14:40 mark of GHS Stretching M51, there is discussion of the desirability of having a negative 'slope' in histogram values to maximize contrast, and how one can use GHS to target a particular region of the histogram to achieve that end.

This seems like a key part of targeted stretching, but I think I'm missing something. Isn't (global) contrast increased by a 'flatter' histogram, as histogram equalization attempts to achieve?

If so, what mental model should I be using w.r.t. increasing the slope of the histogram in order to increase local contrast?

Many thanks,
Craig

BlockHead

I believe my comment was with respect to the histogram as viewed in the logarithmic view. 

I think this is the part you may be missing? 

The way to think about it from a linear histogram is an S-shaped centered at a particular tonal value. This will produce "contrast" at the inflexion of he symmetry point.

-the Blockhead

Craig Kolb

Thanks, Adam, much appreciated. I don't think my confusion was with (semi-)log vs linear space per se, since log is monotonic (i.e., slopes don't change sign, etc.).

I think my misunderstanding was in thinking that the goal is to increase contrast by making the "flatter" part of the histogram less flat per se.

Instead, I believe your point was that you can set the SP to a value of interest, and the resulting stretch will have the effect of making the histogram falloff "smoother" overall. Doing so has the effect of more evenly distributing intensity values over a given range, for example by smoothing the sharp drop in histogram values to the right of the flat section shown in the video. (And of course sharp drops in log-space histogram values are even 'worse' when viewed in linear space...)

In any case, many thanks for the reply, and for all the great content.

Craig