2014-03-24, 13:23
lmyllari I follow what you say about losing <16 and >235 levels when 16-235 is scaled to 0-255, and that banding is introduced by spreading the 220 levels in the Limited source across 255 levels of the Full output.
However shouldn't the scaling be complimentary, so that if you reverse it, the banding is removed - as the initial scale is across a wider range not a narrower range?
Or are you saying that the two scales (LImited->Full and Full->Limited) that in an ideal world we want to avoid - aren't complimentary?
I'm thinking of how the signal would fair through a basic LUT approach and how that would work for the mapping (without any dithering - as I don't believe dithering is usually used to mask the level scaling). Each Limited source level will still map to a unique Full output level, so it should be possible to reverse this with zero loss of original data (other than <16 and >235 that is inevitably losses as it is clipped) removing any banding that has been introduced?
Or am I missing something - or do "real world" implementations make a mess of this?
However shouldn't the scaling be complimentary, so that if you reverse it, the banding is removed - as the initial scale is across a wider range not a narrower range?
Or are you saying that the two scales (LImited->Full and Full->Limited) that in an ideal world we want to avoid - aren't complimentary?
I'm thinking of how the signal would fair through a basic LUT approach and how that would work for the mapping (without any dithering - as I don't believe dithering is usually used to mask the level scaling). Each Limited source level will still map to a unique Full output level, so it should be possible to reverse this with zero loss of original data (other than <16 and >235 that is inevitably losses as it is clipped) removing any banding that has been introduced?
Or am I missing something - or do "real world" implementations make a mess of this?