Highlight Recovery 148 - an analysis & suggestion

Bugs & Suggestions
Post Reply
grubernd
Moderator
Posts: 245
Joined: Thu Jan 12, 2012 8:17 pm
operating_system: Linux
System_Drive: N/A
32bit or 64bit: 64 Bit
processor: from atom to quadcore
Hard_Drive_Capacity: not enough
Location: Graz, Austria
Contact:

Highlight Recovery 148 - an analysis & suggestion

Post by grubernd »

hi there..

3.0.0.148 fixed the pink hue problem pretty much once and for all.
at least in my testing - and my collection of bad raw files from all kinds of cameras is pretty scary.

but.

there is always a but, isn't it?

the new algorithm is so aggressive that it kills contrast and detail down to the midtones. let's take the simple RGB scale for the numbers when we talk about it. 0 being black, 255 being white.
looking at the histogram the new HR affects values way below the 200 mark, which is definitely leaving anything one would consider as a highlight. these are not only soft, fluffy colors, these are colors that have a strong visual impact. especially skintones are really really badly affected by this.

this means that the HR.148 best works with _underexposed_ images, quite the contrary of what one would expect. if you want to know why: underexposed means that the values where the HR is effective are in the 160..255 range. all the things one would consider upper midtones, midtones and shadows are below and need to be brought up.
now there are cameras that work pretty well with heavily underexposed shots, but that is not really the usage where one would need and/or want to use extra HR.

in correctly or overexposed shots, where the general user wants to use HR at it's maximum settings, the HR.148 is reaching down so far that it is pretty much useless.

now what.

well, the main thing i noticed is that the HR Range slider was not changed. this means that this setting adds to the "reaching further down" effect. adjusting the scale of this would help a lot. what is now the lowest setting 1 should in my opinion be somewhere around 30. the 100 could have the same effect as now, but values below 20 should be really limiting the range.

my ideal HR algo would be something that _never_ affects any value below 200. ever. but compresses the current abilities in terms of recovery much more and puts them where the highlights should be: in the most upper parts of the histogram.

this would visually lead to images that have a totally smooth rendition. they should never experience that hard edge when the lights reach 255/255/255 - a value that should be avoided at all costs. anything else is fine, but full white is a value that _always_ asks for trouble and astonishingly is visible on pretty any display, print etc.

TL;DR:
HR in 3.0.0.148 can be improved:
limit the range of affected colors to about 200..255
change the Highlight Range slider so that the "range" of 0..1 becomes 0..30 while keeping 100 at 100.

cheers,
grubernd
use the Monkey to report bugs - include as much precise information as you can provide!
Bibble since 2004 - lots of it in my public archive: http://bilder.grubernd.at
Post Reply