Re: Linear-light RAW 12bit vs R'G'B' 8bit: how much better is it really?
Re: Linear-light RAW 12bit vs R'G'B' 8bit: how much better is it really?
- Subject: Re: Linear-light RAW 12bit vs R'G'B' 8bit: how much better is it really?
- From: Mark <email@hidden>
- Date: Tue, 24 Jul 2007 15:05:55 +0200
Hello Ray, list,
I'm feel like I'm doing progress on this. Thanks for all the inputs,
I hope others around here will profit from these posts too.
There's still something I do not understand:
I understand that CCD sensors work linear in respect to light, and
that the human vision interprets light logarithmically. Now CRT
monitors have a power function curve of approximately 2.5, that is
the main reason why the linear light images coming from sensors get
applied a so called gamma curve, which is a power function of
approximately 0.4. The summed exponentials, which is what the monitor
will emit in the end, has an overall power of approximately 1 - so we
are back to linear light hitting our eyes... getting interpreted
logarithmically.
So the main reason we apply a gamma curve at the CCD images (the
encoder) because of the inherent power function from the CTR
(decoder, I'm not sure how LCDs work, but I believe they mimic the
CRTs gamma through LUTs). There are mostly historic reasons, with
todays computing power it doesn't need to be this way, as proven by
in RAW pictures and even RAW video codecs emerging, think of Cineform
and Red Code.
Now, coincidentally the 0,4 power function is pretty close to a
logarithmic function, so an R'G'B' image is more or less visually
linear, which makes editing it easier as a certain amount of change
affects shadows and brights equally - linear to human perception.
Even more importantly, the 0,4 power function allows to squeeze a
12bit linear image into a gamma corrected 8bit image image in a way
that the shadows get expanded (make use of more bits) and the brights
get compressed (into fewer bits). So in the end, only through the
gamma curve it is possible to get photographic quality in an 8 bit
image - with linear light coding that would not be possible.
That is the part I think I understand.
Confusion starts with prints and film. Film gets projected through a
light bulb and prints reflect light falling in. Naively I would
suppose that both should be coded linearly to light intensity
(transmission or reflection), in order to mimic the real life subject
they represent - as with the example abobe with the R'G'B' image and
the CRT - the overall power function should be 1, so the image gets
interpreted correctly by our eyes.
On the typical press you would get about 20% to 25% dot gain so
that the 50% dot on plate would print as a 70% to 75% dot on
paper. If you plot the input dot area specified and measure the L*
value out on press, you will find that it is visually linear which
means that the end to end gamma is about 2.2.
Well, that breaks my logic described above. I would expect 50% gray
to have L* value of 50 both on print as on film. Why are both encoded
with approx. end to end gamma of about 2.2? Shouldn't overall gamma
always be close to 1 so the light hitting our eyes gets interpreted
correctly.
Cheers
Mark
On 24.07.2007, at 01:16, Ray Maxwell wrote:
Hi Mark,
I will try to explain a little about the word "linear" that is
thrown around a lot in photographic and printing circles. If I am
wrong, I am sure that someone in this group will kindly correct me.
First, the data that comes out of the sensor of your digital camera
is "linear" with respect to energy. That means if twice the energy
comes into the sensor (increase of one f stop or double your
shutter speed) the number that comes out will be twice as big. It
is correct that the first stop at maximum exposure without clipping
contains 2048 levels in a 12 bit system. The next stop down
contains 1024 etc.
Now, the only problem with this is that humans don't react the same
way as sensor chips. Most of the human senses are logarithmic.
This is why we usually covert to a color space with a gamma of 2.2
which is approximately visually "linear". This means that if we
make a change in the highlights of five units and then make a
change in the shadows or mid tones of five units our eyes will
perceive the same amount of change.
This is the reason that sRGB and Adobe RGB 98 are both gamma 2.2.
This makes these spaces visually linear for easy editing.
When everyone was using imagesetters to make film for offset
printing they would "linearize" the imagesetter. This meant that
when you said that you wanted a 50% screen in illustrator the
imagesetter would output a 50% dot on film. When you make a plate
and print on paper you get dot gain. On the typical press you
would get about 20% to 25% dot gain so that the 50% dot on plate
would print as a 70% to 75% dot on paper. If you plot the input
dot area specified and measure the L* value out on press, you will
find that it is visually linear which means that the end to end
gamma is about 2.2. Do to people linearizing their imagesetters
they adopted the term with respect to inkjet proofers when they
usually mean "calibrating" to a known condition rather than making
them linear. The proofer has to emulate the dot gain on press and
would not work if it were "linear".
Hope this helps to understand the difference between energy linear
vs. visually linear.
Ray Maxwell
Mark wrote:
Hi all,
can someone please explain me why 12 bit linear light RAW images
are supposed to be much better than gamma corrected 8 bit images?
Some sources state that 8 bit R'G'B' could be coded linearly with
about 11 bits. So while a 12 bit RAW image does have finer coding
than a 8 bit R'G'B' image, it is not that much more (as one might
naively think at first).
Is that correct or have I gotten it all wrong?
If it is like that, what's the big point in shooting RAW?
Cheers
Mark
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
40dccnet.com
This email sent to email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden