- From: John Bowler <john.cunningham.bowler@gmail.com>
- Date: Wed, 31 Jan 2024 15:16:27 -0800
- To: public-png@w3.org
I cannot parse this sentence in the current draft: >Narrow range images are found in video workflows where the interpretation of sample values below reference black (0% signal level) or above nominal peak (100% signal level). Should "where the interpretation of" be "with"? The base Kodak PhotoCD format does the same thing in an 8-bit encoding; IRC black is '16' and white '235' (but don't quote me). >overshoots and undershoots exist below reference black and above nominal peak in order to preserve processing artifacts caused by filtering/compression If original image data is "normalized" to the range 0..1 then resampled to a sub-pixel offset it will typically produce values above 1 and below 0 and these are not "artefacts" they are correct (you need to use bicubic with the correct coefficient or better to see this happening.) In essence the normalization maps an absolute value >0 to "black" and an arbitrary absolute value to "white" then resampling, particularly with just a pixel offset, recovers the original values at the intermediate locations. It's not that important but the problem is insidious; too many programmers write code that assumes there is some magic in 0.0 and 1.0 >The use of undershoot/overshoot has also been used to preserve additional color volume (both light and color) Indeed. scRGB; what is the transfer function for that? The list of transfer functions seems to be absent. I suggest it needs to be there along with the way to register a new value. I assume values >=128 are private. In any case what is the value for scRGB? Encoding schemes in general: this is not specific to cICP but reference code needs to be in the specification for the various (PQ, HLG) approaches to HDRI (my own is well documented on png-mng-implement, but I'm working on a pure gAMA alternative.) Reference code for cHRM->CIEXYZtriple should also be in there. Referencing ISO or EBU or ITU specs does not help. They are expensive; outside my reach. I'm pretty dubious about the sRGB approach too; the spec was meant to be public but only the WDs were actually released. At least the ICC seems to do it properly. John Bowler <jbowler@acm.org>
Received on Friday, 2 February 2024 04:30:35 UTC