-

Jump to content



Sign up for a free account!

Signing up for an account is fast and free. As a member you can join in the conversation, enter contests and you won't get the popup ads that guests get. Click here to create your free account.

Photo
- - - - -

New Colorization process? Could be interesting to see what it could do to TV releases


This topic has been archived. This means that you cannot reply to this topic.
26 replies to this topic

#1 of 27 Rick P

Rick P

    Supporting Actor

  • 691 posts
  • Join Date: Mar 18 2003

Posted March 10 2005 - 06:45 AM

Don't shoot the messenger, but apparently a new algroithm that seems to be better than any colorization process I've seen...

pictures and video clips here...

http://www.cs.huji.a...s/Colorization/

just FYI

#2 of 27 Gordon St Pierre

Gordon St Pierre

    Stunt Coordinator

  • 150 posts
  • Join Date: Sep 13 2000

Posted March 10 2005 - 06:59 AM

I know the can of worms this will open but I will just say that, That is definately a really amazing process .
Ttiberius

Ttiberius's DVDs

#3 of 27 Rick P

Rick P

    Supporting Actor

  • 691 posts
  • Join Date: Mar 18 2003

Posted March 10 2005 - 07:03 AM

What's more amazing is that they are documenting the algorithm in their paper...

#4 of 27 Lynda-Marie

Lynda-Marie

    Supporting Actor

  • 762 posts
  • Join Date: Jun 03 2004

Posted March 10 2005 - 07:36 AM

What is also interesting is that the colors look natural, rather than an obvious paint job like some colorization processes.

What I've always wondered with colorization, however, is HOW can they determine what the original color of something was?

For example, an old black and white film is colorized, and the lead actress is suddenly wearing a stunning red dress. Was the dress originally red, or is there a way to figure it out, looking at the black and white images?

The shape I'm in you could donate my body to science fiction! - Rodney Dangerfield, "Back to School"

#5 of 27 Scott Kimball

Scott Kimball

    Screenwriter

  • 1,500 posts
  • Join Date: May 08 2000

Posted March 10 2005 - 08:06 AM

Quote:
What I've always wondered with colorization, however, is HOW can they determine what the original color of something was?


They don't... and there lies the problem.

Besides the fact, art design, set design. lighting design and makeup are tuned specifically for black and white photography - not to mention the use of special photographic filters, etc. Even if you knew the colors that were present on the set, you wouldn't want to use them for color output.

-Scott

#6 of 27 Scott Merryfield

Scott Merryfield

    Executive Producer

  • 10,392 posts
  • Join Date: Dec 16 1998
  • LocationMichigan

Posted March 10 2005 - 08:18 AM

Quote:
Besides the fact, art design, set design. lighting design and makeup are tuned specifically for black and white photography - not to mention the use of special photographic filters, etc.

Exactly. So, while the actress may have been wearing a green dress on the set in a particular scene, that color was not selected because it was green, but because it looked a certain way in black and white. Had the movie been filmed in Techicolor, the color choice for the dress may very well have been red (or blue, or whatever).

#7 of 27 James L White

James L White

    Supporting Actor

  • 840 posts
  • Join Date: Jun 29 2002

Posted March 10 2005 - 08:27 AM

Quote:
What I've always wondered with colorization, however, is HOW can they determine what the original color of something was
Easy grass is either green or brown Posted Image

but seriosuly I read soemthing about colorizing movies/TV shows and pretty much the poeple involed in the process just
pick a color for such things as clothes, cars, etc
Samsung HL-S5688W 56" DLP
Panasonic TC-P50X1 50" Plasma TV (bedroom)
Toshiba HD-A2 HD DVD's
Panasonic BMP-BD60
Sony BDP300 Blu Rays (bedroom)

#8 of 27 Jeff#

Jeff#

    Screenwriter

  • 1,942 posts
  • Join Date: Jan 29 2005

Posted March 10 2005 - 09:42 AM

Fortunately the colorization process used for both movies and TV shows is a thing of the past. It was a horrible fad that came and went in the 1980s and insulted the artistic integrity of the original product that was shot in black & white, so I'm not concerned of it ever coming back that way. Posted Image

#9 of 27 Jeff#

Jeff#

    Screenwriter

  • 1,942 posts
  • Join Date: Jan 29 2005

Posted March 10 2005 - 09:42 AM

Double post here -- sorry!

#10 of 27 Rick P

Rick P

    Supporting Actor

  • 691 posts
  • Join Date: Mar 18 2003

Posted March 10 2005 - 09:43 AM

Unless of course you're dealing with something that may have originally been in color and now only b/w prints exist (some of the episodes of Doctor Who come to mind)

#11 of 27 Robert Floto

Robert Floto

    Supporting Actor

  • 739 posts
  • Join Date: Jul 27 1999

Posted March 10 2005 - 10:28 AM

Quote:
Unless of course you're dealing with something that may have originally been in color and now only b/w prints exist (some of the episodes of Doctor Who come to mind)


I would think in that case that a lot of the color can be determined from still photos and those who worked on the episodes.
Movies I would purchase instantly if they were available in their correct aspect ratio:
Death Trap, Innocent Blood, The Shadow, and Remo Williams

#12 of 27 David Galindo

David Galindo

    Screenwriter

  • 1,264 posts
  • Join Date: Mar 30 2003

Posted March 10 2005 - 02:55 PM

That is really amazing...wow. Posted Image

#13 of 27 Brad P

Brad P

    Stunt Coordinator

  • 69 posts
  • Join Date: Feb 10 2005

Posted March 10 2005 - 03:31 PM

Check out what Legend Films has been doing. I'm an opponent of colorization of black and white films, but still - wow.

Examples can be found at legendfilms dot net.

#14 of 27 Lynda-Marie

Lynda-Marie

    Supporting Actor

  • 762 posts
  • Join Date: Jun 03 2004

Posted March 10 2005 - 08:35 PM

Thanks for the info on the colorization process, gentlemen!

Now I understand the request [demand?] Orson Welles made before he passed away, "Keep Ted Turner and his goddamn crayolas away from my movie!" [Citizen Kane]

The shape I'm in you could donate my body to science fiction! - Rodney Dangerfield, "Back to School"

#15 of 27 AndyMcKinney

AndyMcKinney

    Screenwriter

  • 1,915 posts
  • Join Date: Apr 02 2004
  • LocationKentucky, USA

Posted March 11 2005 - 12:46 AM

Quote:
Unless of course you're dealing with something that may have originally been in color and now only b/w prints exist (some of the episodes of Doctor Who come to mind)

Yes, in this instance (not colourising, but re-colouring), colouring B/W material is entirely justified.

Most of the obvious candidates are the scads of British shows that had their colour VTs destroyed and only B/W copies remain (various episodes of Dr. Who, Steptoe and Son, the Goodies and Timeslip, for example), but there's plenty of U.S. telvision like this, too. Some shows that went out live (or almost-live) in colour in the U.S. weren't videotaped: the only backup copies made were B/W kinescopes. The last season (or so) of game show "What's My Line" is one example. And, some shows, such as in the case of "Dark Shadows", are like the previously-mentioned British shows where some episodes have "lost" colour episodes, with only B/W copies remaining.

#16 of 27 CaseyL

CaseyL

    Supporting Actor

  • 803 posts
  • Join Date: Jan 29 2002

Posted March 12 2005 - 03:39 AM

That's a very intruiging process. Cool stuff.

#17 of 27 MatthewLouwrens

MatthewLouwrens

    Producer

  • 3,031 posts
  • Join Date: Mar 18 2003

Posted March 12 2005 - 09:15 PM

Quote:
Unless of course you're dealing with something that may have originally been in color and now only b/w prints exist (some of the episodes of Doctor Who come to mind)
This is already being discussed over at the Doctor Who Technical Forum, where James Insell (one of the main people behind the recently developed RSC process) had this to say:
Quote:
Defining objects through similar brightness levels and edge detection, with a view to flood-filling areas with colour is not a new idea, but I'm pleased to see it being applied and seemingly working so well.

It wouldn't work very well on noisy material. The b/w image examples they are using all seem to be pristine! Also, if it were to be applied to the Pertwee film recordings it would be essential to remove the embedded chroma subcarrier patterning first.
and
Quote:
>>The impression I got was that for the images they used as test subjects, they simply converted some colour images to greyscale and then recolourised them, which might give a pretty artificial result.<<

They have used images that were originated in colour, but this is a good way of contrasting a colourised image with the original. Now, if they also used samples of colour from the original to seed their floodfills with, rather than creating a new colour palette, then that is rather misleading.

>>Does the chroma carrier signal make the greys inconsistent in the images? I'm interested in how this woudl cause a problem. I'm guessing that the software would need clear boundaries between different greys, which simply wouldn't be there in a noisier image?<<

Quite right. And the subcarrier on the film means that adjacent pixels would have drastically different luminance levels.

Films watched in 2005 / 2006 / 2007 / 2008 / 2009 / 2010 / 2011
AFI Top 100 lists: Movies, Thrills - Completed / Laughs - 23 to go
Passions - 39 to go / Heroes & Villains - 10 to go / Songs - 42 to go

#18 of 27 Jeff Jacobson

Jeff Jacobson

    Screenwriter

  • 2,116 posts
  • Join Date: Dec 24 2001

Posted March 13 2005 - 03:43 AM

Matlab code of our algorithm can be downloaded here .

Has anyone tried using this source code themselves? I downloaded it, but couldn't understand the instructions.

Here's what the readme file says:

This package contains an implementation of the image colorization approach described in the paper:
A. Levin D. Lischinski and Y. Weiss Colorization using Optimization.
ACM Transactions on Graphics, Aug 2004.

...

General:

The package provides colorization implementation using two optional
solvers:

1)An exact solver using the matlab "" operator.
This solver is guarantee to provide the global optimum, but running
time can be slow, especially for large images.

2)Multi Grid solver. Fast and efficient, but not always accurate.
This solver is written in C++ and links to matlab as a mex function.

Compilation:

Compiled version for Windows(using Visual C++ 6.0) and Linux are included in this package, but you might need to recompile on your machine.
To compile, type from within matlab:
>>mex -O getVolColor.cpp fmg.cpp mg.cpp tensor2d.cpp tensor3d.cpp


Usage:
You can either use your favorite image editing program
to scribble, or we provide a rudimentary user interface with the function
called (appropriately) cheapUI.m;
Here is an example:
im1=imread('example_res.bmp');
im1=double(im1/255); % images should be double between 0 and 1
cheapUI(im1);

You will see three Matlab figures. The first contains the input
image. You can click on the mouse in this figure to mark points
that define a scribble and then press 'c' to choose a color for
the scribble. You choose a color by clicking in figure 2 on
the color chooser. If the input image is a color image, you
can press 'd' to have the color of the scribble be defined
by the colors in the original image. Pressing 'a' will run
the multigrid algorithm and display results in figure 2.
Pressing 'A' will run the direct solver.

If you prefer to use your favorite image editing program
(e.g. Photoshop, gimp) you need to save two images to
the disk.

1) The original B/W image. The image needs to be saved in an RGB (3 channels) format.
2) The B/W with colors scribbled in the desired places. Use your favorite paint program (Photoshop, paint, gimp and each) to generate the scribbles. Make sure no compression is used and the only pixels in which the RGB value of the scribbled image are different then the original image are the colored pixels.

To run the program define the desired solver and the input file names in the head of the 'colorize.m' file. Then just call the 'colorize' script from within matlab.
An example images pair is included in this package.


#19 of 27 Keith Plucker

Keith Plucker

    Screenwriter

  • 1,007 posts
  • Join Date: Feb 04 1999
  • LocationSacramento

Posted March 13 2005 - 06:56 AM

It looks great but I am bit suspicious. There wasn't a single example of a well known black and white image/video that had the process applied. Every sample they had could have been color originally and converted to black and white for the purposes of their "demo." In other words, it could be a scam. I would like to see the process applied to something that we know is a black and white original, such as the video of the Hindenburg crash.

-Keith
As far as I'm concerned, it's a damned shame that a field as potentially dynamic and vital as journalism should be overrun with dullards, bums, and hacks, hag-ridden with myopia, apathy, and complacence, and generally stuck in a bog of stagnant mediocrity. - Hunter S. Thompson, 1958, from cover letter he wrote for a newspaper job.


#20 of 27 PhilipG

PhilipG

    Screenwriter

  • 2,003 posts
  • Join Date: Jan 13 2000

Posted March 13 2005 - 03:15 PM

Quote:
Has anyone tried using this source code themselves?

It seems to be incomplete. Posted Image You need mex.h, not supplied, and probably Matlab base software too.



Back to TV on DVD and Blu-ray



Forum Nav Content I Follow