Jump to content


Photo

The LUT Process


  • Please log in to reply
9 replies to this topic

#1 Chayse Irvin

Chayse Irvin
  • Basic Members
  • PipPipPipPip
  • 409 posts
  • Cinematographer
  • Los Angeles, CA

Posted 04 January 2007 - 10:47 PM

The LUT Process... Can someone explain it to me as if I were a two year old? Please thank you.
  • 0

#2 Matthew Bennett

Matthew Bennett
  • Basic Members
  • PipPip
  • 38 posts
  • Director

Posted 04 January 2007 - 11:14 PM

I'm sure the scholars on this site will give you a waay better response later, but for now..

A LUT is short for a Look Up Table. In regards to creating images, a Look up Table is used (mostly) to create a linear image from a logarythmic source, ie a TIFF file (linear) from a RAW source (camera RAW).

You would use a LUT to interpret a RAW camera image into a linear form, making an exposure in the process.

I use LUT's in my Andromeda workflow, an upgrade to the DVX100 video camera which allows it to record RAW ccd data. When rendering out the RAW data into a quicktime or a series of stills for editing, I apply a LUT and it grades the image, ie renders it out in a certain coloring and with certain grayscale parameters. Kind of like making a print in photography, I guess.

This is not a simple explanantion and I'm sorry, but if you search for LUT's on this site:: www.reel-stream.com, in the forum, the moderator Juan P. has written many many great and simple explanations of LUTs and why and where you use them as well as the process and theaory behind them.
  • 0

#3 David Mullen ASC

David Mullen ASC
  • Sustaining Members
  • 19759 posts
  • Cinematographer
  • Los Angeles

Posted 05 January 2007 - 01:07 AM

It's a pretty vague term these days. Generally, a Look-Up Table is used to convert one set of image parameters, like color space, so that it looks correct for a certain display technology. But you see the term used a lot in digital intermediates, for example, to describe converting a linear-video HD with its particular color space into a log file with the corrected color space so that it can be recorded accurately to a film negative using a laser recorder.

You also hear the term "viewing LUT", for example, to describe a process that converts the live LOG image off of a Viper or Genesis to look correct on an HD monitor.
  • 0

#4 Chayse Irvin

Chayse Irvin
  • Basic Members
  • PipPipPipPip
  • 409 posts
  • Cinematographer
  • Los Angeles, CA

Posted 05 January 2007 - 02:39 AM

So in "on set" terms, is it a calibration of the monitor or the signal going to the monitor? And if its a calibration of what is going to the monitor is that LUT being recorded? If the LUT is being applied to a HDSDI signal going to a monitor would not the compression process going to the tape change your LUT? (i.e. the changing of the HDSDI 4:2:2 to 3:1:1 as it goes to HDCAM) or is that the point of a LUT... to match your HDSDI signal to your HDCAM, DVCPRO HD, ect., or any painting color corrections applied, recording? Or is it just a LOG-LIN issue and the term only applies to cameras that shoot log? Which cameras shoot LOG aside from Viper and Genesis?
  • 0

#5 David Mullen ASC

David Mullen ASC
  • Sustaining Members
  • 19759 posts
  • Cinematographer
  • Los Angeles

Posted 05 January 2007 - 03:19 AM

Usually a signal going through a viewing LUT process before it reaches the monitor is just for viewing, although I suppose it could also be recorded. But you'd also have the original camera recording that hasn't gone through the LUT.

But I don't think you'd be using a viewing LUT on set for an F900 HDCAM shoot anyway, since the F900 can't produce a RAW LOG signal. What gets recorded by an F900 internal HDCAM VTR (or even if you sent a HD-SDI 4:2:2 signal out to an external HD-D5 or HDCAM-SR deck) already is in the correct linear video color space for HD monitor viewing. Generally LUT's only come into play on an F900/Varicam shoot in the film-out to 35mm. Up til then, everything is handled in linear video from shooting through color-correcting.

Anyway, compression doesn't change the color/gamma characteristics of a video signal.

The Viper, Genesis, Arri-D20, Dalsa and upcoming RED and I assume Silicon Imaging camera output a LOG signal.

The Sony F950 puts out a linear 10-bit 4:4:4 signal, although one can play with the gamma tables to produce a reasonable LOG-ish look, I guess. Digital cameras don't naturally create a LOG signal anyway, especially not one that resembles a 10-bit Cineon Log file, so the term "RAW" is somewhat misleading since the signal has to be processed in some way to make it LOG.

But a LOG image is not viewable as is because it looks very flat, milky, with off-colors on a monitor unless it goes through a viewing LUT. But the term "callibration" is not really accurate -- it's more of a conversion to look correct for the color space of the viewing device.
  • 0

#6 Chayse Irvin

Chayse Irvin
  • Basic Members
  • PipPipPipPip
  • 409 posts
  • Cinematographer
  • Los Angeles, CA

Posted 05 January 2007 - 03:55 AM

I think I'm getting it. What is the difference between a Linear 10-bit 4:4:4 signal and a LOG 10-bit 4:4:4 signal aside from the look. I've seen both examples played back over a screen and they can be very different. But a Linear image changed to look like a Log... what is the inherent difference in the signal? Is a LOG signal by definition a uncompressed, 4:4:4, 10-bit or greater image?

For this coming feature if I were to shoot F900, would it be wiser to attempt to make my image as close as possible to the final look? (like using cinegamma, adjusting the pedestal and certain in-camera options) or would it be wiser to record a flatter image just like a low con neg and allow all color correction to be done later where it can be handled much more gracefully than on set. The problem i see with that is as the F900 records to HDCAM 3:1:1 any alterations I do in post will just add artifacts and degrade the over all image would it not?

PS. Thanks for your help David and Matthew.

Edited by Chayse Irvin, 05 January 2007 - 03:59 AM.

  • 0

#7 Chayse Irvin

Chayse Irvin
  • Basic Members
  • PipPipPipPip
  • 409 posts
  • Cinematographer
  • Los Angeles, CA

Posted 05 January 2007 - 04:29 AM

Another question is how do i ensure my onset monitor is calibrated properly to what the camera is recording? The last shoot I shot Varicam and some stuff came out darker then it looked on set... granted in post I wasn't viewing with the same monitor that was on set. How do I prevent this? Is the HD CRT onset always telling you what is being recorded if no LUT is applied? What if there isnt a CRT but a LCD... if that matters, could a standardized LUT pulled from prep ensure that prep, production, and post monitors are identical? Lastly. How is a custom LUT physically applied to a monitor?
  • 0

#8 Michael Collier

Michael Collier
  • Basic Members
  • PipPipPipPip
  • 1262 posts
  • Gaffer
  • Los Angeles, CA

Posted 05 January 2007 - 07:01 AM

I think I'm getting it. What is the difference between a Linear 10-bit 4:4:4 signal and a LOG 10-bit 4:4:4 signal aside from the look. I've seen both examples played back over a screen and they can be very different. But a Linear image changed to look like a Log... what is the inherent difference in the signal? Is a LOG signal by definition a uncompressed, 4:4:4, 10-bit or greater image?

For this coming feature if I were to shoot F900, would it be wiser to attempt to make my image as close as possible to the final look? (like using cinegamma, adjusting the pedestal and certain in-camera options) or would it be wiser to record a flatter image just like a low con neg and allow all color correction to be done later where it can be handled much more gracefully than on set. The problem i see with that is as the F900 records to HDCAM 3:1:1 any alterations I do in post will just add artifacts and degrade the over all image would it not?

PS. Thanks for your help David and Matthew.



Log and Linear are just methods of giving value to the binary numbers. Linear gives the same weight to each increase in the root bit. So ( 4 bit binary example) 0000 = 0, 0001 =1, 0010 = 2 0011 = 3, 0100 = 4, 0101 = 5, etc. So the relative lumanence value (lum of the color channel in question) is directly related to the binary value.

Log does not increase its relative valuation at a constant rate. If you were to plot a graph of the valuation (where binary value 0-1024 (10 bit) is on the horizontal, and lumenance value is on the verticle) Linear would plot as a straight diagonal line, while log would plot as a parabolic curve.

A LUT is nothing more than a computer cheat sheet. Imagine if someone gave you the multiplication table and started asking a bunch of multiplication questions. You wouldn't do any math, just find look at the table and find the answer.

Calculating the change from log to lin takes more time than is needed. A LUT has a row of input values and a row of output values. There are only 1024 different input possibilities (10 bit), so why not calculate them all at once, and have the answer ready as soon as it can be accessed? This proccess applies in several areas of digital tech. DIs (from what I read, never done one) use LUTs to compensate for the response of the film being recorded to. That way the relative color shown on screen will match the films output.

LUTs also find their way into just about every other electronic device. Older video cameras used LUTs to handle white balancing of the image, to save on proccessing, memory and power requirements (newer ones have more complex proccessing done in real time with more effiecent cheaper units) Video games, electronic calandars, even your microwave probably has one in its microchip.

As for the compression being affected by the LUT, it is one generation loss. If your doing any sort of post you loose at lease one generation. When LUT are applied, they always are proccessed uncompressed. Most NLEs do all internal proccessing at high uncompressed bit rates, eliminating any internal errors to affect the video. Using a LUT at the time you conform the original footage for grading will not adversley affect the quality of your video (no more than would happen anyway).

From what I understand of the DI proccess, a lot of times its just applied to the output monitor only, while the uncorrected data gets recorded to film.



Another question is how do i ensure my onset monitor is calibrated properly to what the camera is recording? The last shoot I shot Varicam and some stuff came out darker then it looked on set... granted in post I wasn't viewing with the same monitor that was on set. How do I prevent this? Is the HD CRT onset always telling you what is being recorded if no LUT is applied? What if there isnt a CRT but a LCD... if that matters, could a standardized LUT pulled from prep ensure that prep, production, and post monitors are identical? Lastly. How is a custom LUT physically applied to a monitor?



Your best bet is to use a proffesional CRT monitor to view your images, and have it properly set up against a chip chart. This can be time consuming, and it won't ever be perfect. Variations in peoples eyesight, changes in tempature, sudden jars to the monitors, solar flares, high power lines, alien spacecraft can all affect the setups accuracy. At a certain point you should lean heavily on the waveform/vect. That will give you the best idea of what the data looks like. If you want a certain corner to be dark, you can see just how dark it is. Getting to know the waveform is definatley a good idea. Using a well calibrated monitor and the waveform is a great combo.
  • 0

#9 Otis Grapsas

Otis Grapsas
  • Basic Members
  • PipPip
  • 44 posts
  • Digital Image Technician
  • Athens, Greece

Posted 21 January 2007 - 06:40 AM

In digital signal processing a lookup table is a precalculated result of a y=f(x) function. Instead of going through calculation of f(x) for each x, you just pick a value precalculated for your x and stored in memory. When the process asks for f(4) it is pointed to a memory location that stores the value of f(x) for x=4.

In small precision systems like integer 8 and 10bit, all values are precalculated since not much memory is needed to store the results. This is as good as calculating f(x) in terms of precision.

In high precision systems like 24bit or floating point ones, some values are precalculated and others are calculated using the nearest higher and lower precalculated values using a simple algorithm. This is called interpolation. This is inferior in terms of precision compared to a straight calculation of f(x).

A LUT table in an imaging system can do many things.

It can map the CCD output. CCD pixels produce voltage for each photon they get and this is inherently linear as a process. 10x as many photons is 10x as much voltage to the AD converter. This does not correspond to the way humans see things so we need to map this linear format to log. Human perception of brightness is logarithmic to luminance but CCD coding of brightness is linear to luminance. the linear coding of CCDs wastes most of its precision in highlights so in order to have good shadow precision you need a higher bit resolution in linear format compared to log format. An 8 bit video format needs a CCD with a 10bit or better converter. This LUT in its simplest form is called a gamma curve.

It can create various forms of nonlinearity in the system that allows a low resolution 8 bit format to fit more latitude from the usually more capable imaging system. You create a function that will map lets say 4bits of latitude in highlights to 2bits of digital image. You have 10bits of latitude in an 8bit format when you do this. But less tonal resolution in highligths, but this another story. This variation build on the simple gamma curve is usually called a cine curve or a black stretch or various other marketing terms.

Some systems allow you to work with linear image formats. There is no precision benefit to this, even if the precision of the linear format appear a little higher like 8 bit log compared to 10 bit linear. There is one benefit though. You can keep some highlight detail in exchange for some shadow detail when you convert to 8bit, by varying the LUTs used. This might be useful in some situations.

A LUT can also map the digital output from the 3 colors on a CCD in order to compensate for different color temperatures in lighting. This is usually implemented using different gain on the three channel for getting most of the desired effect though.
  • 0

#10 Otis Grapsas

Otis Grapsas
  • Basic Members
  • PipPip
  • 44 posts
  • Digital Image Technician
  • Athens, Greece

Posted 21 January 2007 - 07:14 AM

This LUT in its simplest form is called a gamma curve.


I'm not refering to gamma correction here. Gamma correction is a mapping to compensate for display nonlinearities. You gamma correct something so that it will display properly on a CRT or other nonlinear display.

LOG conversion in the other hand is just a prerceptual mapping. A way to make better use of the storage bits for encoding for human vision.

Edited by Panagiotis Grapsas, 21 January 2007 - 07:16 AM.

  • 0


Ritter Battery

CineTape

The Slider

CineLab

Opal

Willys Widgets

Glidecam

Rig Wheels Passport

Tai Audio

Aerial Filmworks

Visual Products

Technodolly

Broadcast Solutions Inc

Abel Cine

rebotnix Technologies

FJS International, LLC

Wooden Camera

Paralinx LLC

Media Blackout - Custom Cables and AKS

Metropolis Post

Gamma Ray Digital Inc

FJS International, LLC

Paralinx LLC

CineTape

Abel Cine

Media Blackout - Custom Cables and AKS

CineLab

Aerial Filmworks

The Slider

Opal

Wooden Camera

Gamma Ray Digital Inc

Metropolis Post

Visual Products

rebotnix Technologies

Ritter Battery

Technodolly

Glidecam

Willys Widgets

Rig Wheels Passport

Tai Audio

Broadcast Solutions Inc