Why not linear processing as default in edition?

[Guillermo Luijk]

Linear edition has many advantages vs gamma edition. The only reason I find today to stay working on gamma corrected images is if the image is encoded as 16-bit (or less) integer levels, where a linear encoding would mean danger of posterization in the shadows.

But what about a 32-bit floating point image editor? why PS remains integer 16-bit (15-bit actually) for most of its functions? would images be so memory hungry? or hard to process with present computers? is it just not to having to redesign the present edition tools? when a fully functional 32-bit floating point version of PS with all its features adapted to properly work with linear images?

Exposure changes with no change in Hue, easy logarithmic histograms with EV divisions, curves and other tools adapted to work in terms of input vs output f-stops (a very photographic approach),...

[madmanchan]

At this point the main reasons are legacy code and speed, particularly the former. Photoshop having been around for a long time, it has a ton of code in it organized around 8-bit and 16-bit fixed-point image processing. Aside from being time-consuming to replace it with a full 32-bit processing pipeline, keep in mind that doing so would immediately break most existing PS plug-ins ...

[Chris_T]

Linear edition has many advantages vs gamma edition. The only reason I find today to stay working on gamma corrected images is if the image is encoded as 16-bit (or less) integer levels, where a linear encoding would mean danger of posterization in the shadows.

But what about a 32-bit floating point image editor? why PS remains integer 16-bit (15-bit actually) for most of its functions? would images be so memory hungry? or hard to process with present computers? is it just not to having to redesign the present edition tools? when a fully functional 32-bit floating point version of PS with all its features adapted to properly work with linear images?

Exposure changes with no change in Hue, easy logarithmic histograms with EV divisions, curves and other tools adapted to work in terms of input vs output f-stops (a very photographic approach),...
[{POST_SNAPBACK}][/a]

By "linear editing", are you referring to something like this?

[a href=\"http://www.aim-dtp.net/aim/evaluation/gamma_error/index.htm]http://www.aim-dtp.net/aim/evaluation/gamma_error/index.htm[/url]

DISCLAIMER: I'm no expert on this topic, but very curious about it. The above site/author is known to be controversial, which makes it interesting at this stage of digital imaging.

[bjanes]

At this point the main reasons are legacy code and speed, particularly the former. Photoshop having been around for a long time, it has a ton of code in it organized around 8-bit and 16-bit fixed-point image processing. Aside from being time-consuming to replace it with a full 32-bit processing pipeline, keep in mind that doing so would immediately break most existing PS plug-ins ...
[{POST_SNAPBACK}][/a]

It would appear that Photoshop and ACR are victims of their own success. Already there are specialized raw converters such as Raw Magick which use floating point processing and claim to have better demosaicing with less Moiré as shown in the comparison in the link below:

[a href=\"http://www.rawmagick.com/comparisons2.html]Raw Magick Comparisons[/url]

The features that Guilermo listed would be most welcome.

Bill

[madmanchan]

I doubt that improved demosaicing algorithms benefit from a 32-bit floating point image processing pipeline. The most important place to have higher precision would be in shadow tonal adjustments since during raw conversion there is usually a curve applied with a high slope in the shadows.