White level and contrast display vs mastering and video dynamic range? - Page 2 - AVS Forum
Forum Jump: 
Reply
 
Thread Tools
post #31 of 40 Old 06-13-2011, 04:02 PM
ADU
AVS Special Member
 
ADU's Avatar
 
Join Date: Apr 2003
Posts: 6,333
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 96 Post(s)
Liked: 153
^ Still workin on a response to this (and makin a couple more visual aids to assist with that). I think you make an interesting point. If CCD sensors are linear and the display technology is linear, then why not just pass the linear data straight on through?

ADU
ADU is offline  
Sponsored Links
Advertisement
 
post #32 of 40 Old 06-13-2011, 04:40 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,621
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 28 Post(s)
Liked: 167
Quote:
Originally Posted by ADU View Post

^ Still workin on a response to this (and makin a couple more visual aids to assist with that). I think you make an interesting point. If CCD sensors are linear and the display technology is linear, then why not just pass the linear data straight on through?

The big reason why they haven't gone linear yet is due to bit depth.

8bits is not enough to do linear data, and until a few years ago the processing power to handle more bits wasn't price effective.

I'm sure the future will likely bring linear encoding.

darkest possible shade above black

8bit gamma 2.3
0.000002917 Yn

8bit linear
0.003921

10bit linear
0.0009775

12bit linear
0.0002442

18bit linear
0.000003814

So you are looking at moving over double the amount of bits around to get similar dynamic range.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is offline  
post #33 of 40 Old 06-13-2011, 05:29 PM
AVS Special Member
 
dlarsen's Avatar
 
Join Date: Mar 2001
Location: Beaverton, OR, USA
Posts: 1,908
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 11
There's also the 800lb Gorilla in the corner with the "Legacy" tattoo. He's there in all of our present displays and standards like Rec.709, SMPTE, sRGB, etc.

We still tote around legacy issues like interlace, chroma subsample, BTB/WTW reserved codewords, etc, etc. I'd be happy to see all these notions die too.
dlarsen is offline  
post #34 of 40 Old 06-13-2011, 09:06 PM
AVS Special Member
 
Mr.D's Avatar
 
Join Date: Dec 2001
Location: UK
Posts: 3,307
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
Quote:
Originally Posted by sotti View Post
The big reason why they haven't gone linear yet is due to bit depth.

8bits is not enough to do linear data, and until a few years ago the processing power to handle more bits wasn't price effective.

I'm sure the future will likely bring linear encoding.

.
Actually its more to do with the processors not being able to handle float calculation for linearising things like 10bit log film scans which when linearised have values above 1 and under 0. You can pretty much handle any imagery currently created by man with 14bits ( in practice Kodak say its more like 20bits for linearising a 10bit log film scan but I suspect they are slcing their onion a little too finely).

That aside its routine to work in 32bit float linear these days for material sourced from video , digital SLR , RED, Film , CGI you name it. . The transform curves for different colorspaces are just baked in according to delivery.

Eventually it will all be linear material with linear displays. Essentially the colorspace of the real world and handily the colorspace of 3d lighting models in CGI.

digital film janitor
Mr.D is offline  
post #35 of 40 Old 06-15-2011, 05:28 PM
ADU
AVS Special Member
 
ADU's Avatar
 
Join Date: Apr 2003
Posts: 6,333
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 96 Post(s)
Liked: 153

Some graphical examples of stuff discussed above.

Images in this post (and the one that follows) were designed to appear correct on an ~2.45 gamma display. They should appear approximately the same on a 2.2 monitor, though the actual luminance measurements of the color swatches will be slightly off their proper values on a 2.45 display. To best see how these patterns would appear in a dim surround typical of home viewing, switch the AVS forum "skin" to Black Mode in your User Profile, and if possible turn contrast up a bit on the display or dim the surrounding illumination.

On an ~2.45 gamma display, L* 50 would correspond to a 50% stimulus, or RGB Code 127 in full 0-255 swing levels as shown here...
 

 
L* Value
(% Lightness)
 
  000 025 050 075 100  
L*
Average Surround
 

 

 

116 ( .1842 relative luminance ^ .3333 ) - 16 = 50 L*

.50 stimulus ^ 2.45 "gamma" = .1830 relative luminance

 

As you can see from these two equations, the relative luminance of L* 50 is virtually identical to a 50% stimulus on a 2.45 display.

As mentioned previously though, L* represents lightness perception in an average surround. So the grays in the pattern above should appear most uniform in an average surround rather than the dim to dark surround typical of home viewing.

To approximate perception of lightness in a dim surround, all one need do is apply an ~1.225 correction to the L* pattern above as shown here...
 

 
L* Value
(% Lightness)
 
  000 025 050 075 100  
L*
Average Surround
 
L* Corrected for
Dim Surround
(L* DIM)
 

To simplify things, I'll refer to the corrected pattern as "L* DIM".

You can see that the difference between the two (L* and L* DIM) isn't all that much. But in the dim to dark surround typical of home viewing, the grays in L* DIM should appear more uniform or equally spaced apart in lightness. The difference is probably most noticeable in the step between 0 (black) and 25 (dark gray). In a dim surround, those two colors should appear a bit too far apart in lightness in the L* pattern.

The L* DIM pattern represents approximately the ideal perceptual coding for a dim to dark surround typical of home viewing... So let's see how it compares to a 25%, 50% and 75% stimulus on some different displays.

First, a 3.0 gamma display...
 

 
% Lightness (L* DIM)
% Stimulus (Display)
 
  000 025 050 075 100  
L* DIM  
3.00 Gamma Display  

As you can see above, the brightness of a 25%, 50% and 75% stimulus on ~3.0 gamma display is an excellent match to the ideal steps in lightness in L* DIM. So the coding on a 3.0 display would be just about optimal for a dim surround from a perceptual standpoint. For an image to appear "correct" though on a 3.0 display, it would also need to be encoded with gamma closer to .41, rather than the ~.50 encoding in Rec. 709.

However, a 2.45 gamma display (best for ~.50 encoding imho) is also a pretty good match to L* DIM...
 

 
% Lightness (L* DIM)
% Stimulus (Display)
 
  000 025 050 075 100  
L* DIM  
2.45 Gamma Display  

A 1.0 (linear) gamma display though is a poor match...
 

 
% Lightness (L* DIM)
% Stimulus (Display)
 
  000 025 050 075 100  
L* DIM  
1.00 Gamma Display  

On a linear display there's a much larger perceptual jump in lightness between the 0% and 25% stimulus.

Cont'd next post...


ADU
ADU is offline  
post #36 of 40 Old 06-15-2011, 05:32 PM
ADU
AVS Special Member
 
ADU's Avatar
 
Join Date: Apr 2003
Posts: 6,333
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 96 Post(s)
Liked: 153

Cont'd from above...

That "perceptual disconnect" on a linear display gets progressively worse as the codes get closer to black (as indicated in the 0-25% range below), demonstrating that the distribution of codes is more efficient on a 2.45 or 3.0 display.
 

 
% Lightness (L* DIM)
% Stimulus (Display)
 
  000 005 010 015 020 025  
L* DIM  
1.00 Gamma Display  

As others mentioned, a linear system would require about double the bit-depth (~16 bits) to produce shading in the darker coding range comparable to our current nonlinear 8-bit system. And many of those add'l bits would be wasted on perceptually insignificant information in the higher luminance range.

For dim surround viewing, a "linear system" would probably also still need a nonlinear component (of ~1.225) somewhere in the chain to provide "picture rendering". So the data would probably still need massaging somewhere, or home theaters might also go the way of the dodo... no?

If you're working with linear data, and simply trying to preserve rendering intent in our current nonlinear paradigm, I think the best way to accomplish that is to color correct the linear data on an ~1.225 lutted display (if you have a display that can do that without producing unpleasant shading artifacts), and then simply zap the data with a square root (aka .50 gamma correction) when converting to 8-bit. If everything else (especially the ratio of ambient light to white level) is equal, you should get virtually the same image on an ~2.45 (or 2.2 - 2.5) display.

 

1.0 linear encoding * 1.225 display gamma = 1.225 screen gamma

.50 nonlinear encoding * 2.45 display gamma = 1.225 screen gamma


ADU
ADU is offline  
post #37 of 40 Old 06-15-2011, 09:05 PM
AVS Special Member
 
Mr.D's Avatar
 
Join Date: Dec 2001
Location: UK
Posts: 3,307
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
Quote:
Originally Posted by ADU View Post


If you're working with linear data, and simply trying to preserve rendering intent in our current nonlinear paradigm, I think the best way to accomplish that is to color correct the linear data on an ~1.225 lutted display (if you have a display that can do that without producing unpleasant shading artifacts), and then simply zap the data with a square root (aka .50 gamma correction) when converting to 8-bit. If everything else (especially the ratio of ambient light to white level) is equal, you should get virtually the same image on an ~2.45 (or 2.2 - 2.5) display.

1.0 linear encoding * 1.225 display gamma = 1.225 screen gamma

.50 nonlinear encoding * 2.45 display gamma = 1.225 screen gamma



Displaying linear images is not problematic whatever the end display paradigm. That's not the point of it. Linear allows easier and arguably more accurate color maths operations (decoupling gamma from gamut for example) and makes filtering operations more accurate. As I've said 90% of the professional material worked these days is linear right up till delivery. CGI is inherently linear because of the linear nature of 3d lighting modeling systems.

Also if we move to HDR imagery as a format in the future ; again arguably a much more effective upgrade of current consumer video that merely throwing more pixes at it, we will need a linearrendering intent to use it effectively and relevantly and to make it easily backwards compatable with legacy non-linear formats.

Also just for note. Film scans are essentially HDR . They contain headroom and footroom and to map them into video decisions have to be taken as to how much of these areas to transfer. The simple maths operations you describe for converting linear to video only work if the material itself was video in the first place. What with the arbitrary customisation that can happen on digital cameras in terms of the exposure and the inherently HDR nature of film negative AND the prevalaent paradigm in film to overexpose on capture to place the recording of black variation above the toe of the film , the simple maths is never enough to turn this into video , you still need a human being to eyeball it. The numbers never quite work, alos explains why you never get exactly the same video transfer from two different people.

The bit expenditure will very swiftly not be the over-riding issue. Legacy material can be linearised at source realtime with a simple LUT.

digital film janitor
Mr.D is offline  
post #38 of 40 Old 07-23-2011, 01:54 PM
ADU
AVS Special Member
 
ADU's Avatar
 
Join Date: Apr 2003
Posts: 6,333
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 96 Post(s)
Liked: 153
Quote:
Originally Posted by Mr.D View Post

Displaying linear images is not problematic whatever the end display paradigm. That's not the point of it. Linear allows easier and arguably more accurate color maths operations (decoupling gamma from gamut for example) and makes filtering operations more accurate. As I've said 90% of the professional material worked these days is linear right up till delivery. CGI is inherently linear because of the linear nature of 3d lighting modeling systems.

Also if we move to HDR imagery as a format in the future ; again arguably a much more effective upgrade of current consumer video that merely throwing more pixes at it, we will need a linearrendering intent to use it effectively and relevantly and to make it easily backwards compatable with legacy non-linear formats.

You make some interesting points here. I can see advantages to physical (ie linear) coding on the front end for video production, and possibly also on the back end for some gaming and virtual reality apps. However, I think there are (and probably always will be) advantages to both physical and perceptual coding. The advantages to linear coding are readily apparent from a physical modeling standpoint. But it's not as intuitive in terms of graphic design, or as efficient for distribution. So perceptual coding is really more advantageous in these cases.

For example, I think it's easier for most people to conceive of middle gray as being at or near the middle of the coding range, ie around a 50% stimulus rather than 18% luminance, even though a perceptually middle gray would be somewhat darker than these values on a 2.2 to 2.5 display in the dim to a dark surround conditions typical of home video.

The same principles apply to HDR as to regular video. If you need 10 or 12 bits to get a decent looking HDR image with nonlinear coding, then you'd need about 20 to 24 bits (per color component) to get similar performance in a linear HDR system.

Nonlinear content scales to wider ranges of contrast better than linear content. Try raising the contrast on your monitor and you'll see what I mean in the sample below...
 

% Stimulus: 00 05 10 15 20 25
1.00 (Linear) Gamma Display
2.45 Gamma Display
3.00 Gamma Display

The steps on the hypothetical 2.45 and 3.0 nonlinear displays above remain relatively even, while the gap between 0% and 5% stimulus appears to grow every wider on the linear display. The greater the contrast is on the linear display, the more bits will be needed to fill that gap.

There is nothing inherently incompatible between HDR and our current nonlinear 8-bit system btw... if it's used properly. Additional bits would just give a somewhat smoother-looking result on higher than normal contrast displays.

(Note: Luminance values in the patterns above are adjusted for an ~2.45 gamma display, just like the other patterns on this page.)


ADU
ADU is offline  
post #39 of 40 Old 07-23-2011, 11:16 PM
AVS Special Member
 
Mr.D's Avatar
 
Join Date: Dec 2001
Location: UK
Posts: 3,307
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
Quote:
Originally Posted by ADU View Post

There is nothing inherently incompatible between HDR and our current nonlinear 8-bit system btw... if it's used properly. Additional bits would just give a somewhat smoother-looking result on higher than normal contrast displays.

(Note: Luminance values in the patterns above are adjusted for an ~2.45 gamma display, just like the other patterns on this page.)

Linear HDR imagery is commonly passed about at 32bits (usually as Open EXR files). Its processed at 32bit float.

Pretty much all the points you raise regarding perceptual encoding are only in existence as mechanisms to deal with bit starvation, they are essentially compression techniques.

In terms of working with imagery "intuitively" all common image operators ( adds , mults, divides you name it) give more predictable results right across the intensity range whilst operating with linear imagery. This is a fundamental point in digital imagery. If you look at something like "Nuke" for example its primary mode of operation converts all incoming material into linear and then reconverts on output according to image requirements.

The display issue scaling doesn't really apply if you have a truly linear display being fed linear imagery. In practice you can have imagery look exactly the same ( well down to about 4 decimal places) on a video display as video with the linear image exactly the same on a linear display next to it.

digital film janitor
Mr.D is offline  
post #40 of 40 Old 05-07-2013, 07:07 PM
ADU
AVS Special Member
 
ADU's Avatar
 
Join Date: Apr 2003
Posts: 6,333
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 96 Post(s)
Liked: 153

Edited some of my posts in this thread to fix some formatting issues caused during the AVS website changeover.

 

Also, the ITU recently approved a new standard for display gamma in HDTV production (BT.1886) with an exponent of 2.4.


ADU
ADU is offline  
Reply Display Calibration

User Tag List

Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off