FujiFilms WBS solution.

Started Nov 29, 2011 | Discussions thread
Shop cameras & lenses ▾
Mark H
Veteran MemberPosts: 3,205
'Gamma correction' (no pun intended)...
In reply to Kim Letkeman, Dec 8, 2011

Kim Letkeman wrote:

Mark H wrote:

Kim Letkeman wrote:

Remember, RAW data is stored logarithmically by stops and is one 14-bit number.

RAW data is linear , not 'logarithmic'.

Yes, a nit correctly picked.

Yet another error you describe as 'a nit' - I guess, some people just seem to be more prone to 'nits' than others.


Amazingly, even that 'Bruce Fraser' Adobe article contains a significant misleading inference...

'Gamma' (more correctly 'gamma-correction') is not there to correct/compensate for the logarithmic nature of human vision - gamma correction is actually there to correct/compensate for the non-linear response of CRT displays.

CRT displays tubes respond non-linearly (an exponential power response) - their brightness is non-linear wrt the input signal (voltage applied to their electron-guns).


Image data is consequently ' gamma-encoded ' to compensate for the CRT response - there by making the ' total system ', from the sensor image capture to the image display, more linear .

Modern flat panel displays are designed to have the same response characteristic as CRT monitors, largely because of the 'legacy' of all the pre-existing hardware and image data.

It might seem strange to persist with a practise designed to overcome a problem of an archaic, near obsolete, technology - but as it happens, the gamma-correction doesn't actually hurt because it mostly just compresses/looses the unnecessarily/excessively fine tonal gradation in the higher/brighter part of an image capture by a linear image sensor.

So the suitability of 'gamma correction' to the nature of human vision, is actually little more than a 'happy accident, or coincidence' resulting from the completely opposite CRT response characteristic.

The sum, of the non-linear CRT response with that of the image gamma-correction, results in a nominally ' linear reproduction system - so we 'see' nominally the same tonal reproduction as the original subject/scene (not withstanding 'stylisations' to simulate 'film' etc).

A 'side effect' is that this disposes of/discards much of the excessive number of highlight levels capture by the linear sensor - but it also stretches the lower 'noisier' shadow levels further apart, which is the primary reason why camera A/D converters have many more 'bits' than actually might appear to be needed for 24 bit (8 bit per colour) images.

Linear gamma ... meaning that each stop uses half of the remaining values available. Which is what I meant. Top stop gets 1/2 the values, next stop gets 1/4, next stop gets 1/8, then 1/16, then 1/32 and finally 1/64.

Except you clearly said "...RAW data is stored logarithmically by stops..." ...which really couldn't be more wrong.

Even viewing 'by stops' - then the RAW data actually would appear as 'inverse logarithmic' (exponential) - and actually needs a logarithmic function to be applied in order to convert it to appearing evenly distributed 'by stops'.

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow