Chori Chori 1956 - three songs in colour.

[Sukesh]

yeh raat bheegi bheegi

http://www.youtube.com/watch?v=WNPAABBkk7M

aajaa sanam

http://www.youtube.com/watch?v=CzwESL4uSLE

jahaa.n mei.n jaati hoo.n

http://www.youtube.com/watch?v=UFkpDKWmpnQ

Enjoy
Sukesh

[Abhay Phadnis]

Robert Frost wrote:

What comes over a man, is it soul or mind---
That to no limits and bounds he can stay confined?
...Why is his nature forever so hard to teach
That though there is no fixed line between wrong and right,
There are roughly zones whose laws must be obeyed?

He was writing about something quite different, of course, but he could well have been writing about this lamentable urge to tamper with well-loved things instead of leaving them as they are!

Warm regards,
Abhay

[AR]

The picture quality in most old films was poor. This project to both
restore and colorize the dull black and white frames needs to be
encouraged. And it's not as if the end result has any perceptible
artifacts. In fact, digital post-processing of old films has become
quite an active area of research within the field of digital image
processing and computer vision. I have no idea how exactly the
restoration was achieved in India, but there is one very successful
research group in Ireland which focuses on such applications:

http://www.mee.tcd.ie/~sigmedia/People




> Warm regards,
> Abhay

[Abhay Phadnis]

On Sunday, 29 April 2012 00:52:34 UTC+5:30, AR wrote:
(snip)

>
> The picture quality in most old films was poor. This project to both
> restore and colorize the dull black and white frames needs to be
> encouraged.

Restoring a print and "colorizing" it are two completely different things. Provided it is done with proper care and respect for the original creation, restoration may indeed be welcome. Colorization, on the other hand, is tampering with the original creation: it adds no discernible value. (All IMO, of course!)

Warm regards,
Abhay

[AR]

Colorization can be considered one form of restoration. Noise, blur,
flicker are annoying artifacts, but a gray-scale image can itself be
considered an artifact (less annoying perhaps) as most people see the
world in color.

Producing a remixed version of an old song with changed orchestration
or percussion may be considered tampering, especially if it tends to
alter or spoil the mood of the original song. I don't see how
introducing color in a gray-scale image frame can be considered
tampering, unless there are perceptible color artifacts like sudden
changes in hue or intensity. Or if I were to process the frames of the
movie in such a way that they appeared to look like images from a
cartoon strip. Just wondering.



> Warm regards,
> Abhay

[vrk]

totally agree with you.

recently read that scorcese has restored (digitally) uday shankar's 1948 film kalpana.

any such effort has to be encouraged, because our old films were technically rotten.

[vrk]

although i do wonder why our researchers do not restore the film in black and white format itself. why does it always have to be in colour. i am sure there is a lot of scope of improvement even in black and white.

i remember seeing a dvd of roman holiday (which was digitally restored) and the making film explained the difference. I was amazed by what i saw and wondered why the same wasn't done to pyaasaa and kaagaz ke phool (especially the latter)
On Sunday,

[Abhay Phadnis]

On Sunday, 29 April 2012 02:35:07 UTC+5:30, AR wrote:
(snip)
>

> Colorization can be considered one form of restoration. Noise, blur,
> flicker are annoying artifacts, but a gray-scale image can itself be
> considered an artifact (less annoying perhaps) as most people see the
> world in color.

If you do not consider black-and-white cinema an art form in itself - indeed, if you see it as a shortcoming that needs to be overcome by adding colour to it - then I guess we just have very different views on the subject! I am happy to stick to the "gray-scale" image - I don't see it as an "artifact" that needs fixing.

>
> Producing a remixed version of an old song with changed orchestration
> or percussion may be considered tampering,

So you consider the aural component as inviolate but are not willing to extend the same courtesy to the visual component!

Warm regards,
Abhay

[Abhay Phadnis]

On Sunday, 29 April 2012 02:35:07 UTC+5:30, AR wrote:
(snip)
>

> Colorization can be considered one form of restoration.
>

Sorry to respond piecemeal, but I forgot to add that this is incorrect: restoration (especially as it is used in the arts) means bringing something as close as possible to its original form. Adding colour to a black-and-white film, therefore, is not "restoring" it.

It is also worth noting that restoration in any of the arts is almost always a contentious topic. The controversy over the "restoration" of the Ajanta paintings is a good case in point.

Warm regards,
Abhay

[Vijay]

I am in total agreement with Abhay. Colorisation is an answer to a
question no one had asked. I can't imagine ever enjoying my favourite
B&W movies in their colorised versions (I have tried watching a
couple, It's a Wonderful Life the more recent one (Comes as a standard
extra in the Blu Ray package); couldn't get past 5 minutes).

I do have an issue with overenthusiastic digitisation of Blu Ray/DVD
transfers where the grain/noise gets completely removed, even where
the director had deliberately used grain, say, to picturise an
interior or a night scene. Casablanca (and many other Warner Brothers
earlier High Definition transfers) suffer from this. It can take on a
very artificial look. (A very blatant example of this is the Blu Ray
transfer of Om Shanti OM, where all grain/noise has been removed and
the whole film carries a very waxed/polished look). Criterion
transfers are excellent in this respect where digital enhancements are
limited to clearing up artefacts that may have cropped up because of
the age of the print, but preserving the grain/noise that was germane
to the original version.

Best regards,

Vijay Kumar

[AR]

On Apr 29, 2:45 am, Abhay Phadnis <aphad...@gmail.com> wrote:
<snip>>

> > Producing a remixed version of an old song with changed orchestration

> > or percussion may be considered tampering, ***especially if it tends to

alter or spoil the mood of the original song. I don't see how
introducing color in a gray-scale image frame can be considered
tampering, unless there are perceptible color artifacts like sudden
changes in hue or intensity. Or if I were to process the frames of the
movie in such a way that they appeared to look like images from a

cartoon strip. Just wondering.***

>
> So you consider the aural component as inviolate but are not willing to extend the same courtesy to the visual component!
>

I think you didn't quote (read?) my post in its entirety :-).

I didn't say audio was inviolate. I said I consider a remix as
tampering because many of them change the orchestration, laya, etc.
completely and thereby the mood of the song. One such example (amongst
hundreds) is the remake of Shamshad's "saiyaan dil mein aana re" -
with very different (and loud) orchestration, and sung by a singer
whose pronunciation was terribly anglicized. Generously sprinkled with
English phrases all over, and picturized on girls with cyan-colored
lips, this was a "tampered" product. However, HMV's revival series
was different from that. Those songs were remastered - recorded again
in a modern studio with the same instruments as the original, and
overlaid onto the original voice track. I hope you see the difference
between these two cases.

Adding color to a gray-scale movie does not change the mood of the
movie. Or transform it in such a way that it looks like (say) a
cartoon film. That's why it's not tampering.
I wouldn't appreciate a beautiful Himalayan scenery or a scene of
orchards or flower-beds if the image was in a grayscale - the numerous
hues of the flowers and the grass and the fruits would be missing! I
wouldn't have appreciated the picturization of "pyaar kiya to darna
kya" as much, had it been grayscale.

Should it be any surprise that modern film-makers don't make black and
white movies any more?



> Warm regards,
> Abhay

[Abhay Phadnis]

On Monday, 30 April 2012 07:57:48 UTC+5:30, AR wrote:
(snip)

> I didn't say audio was inviolate. I said I consider a remix as
> tampering because many of them change the orchestration, laya, etc.
> completely and thereby the mood of the song. One such example (amongst
> hundreds) is the remake of Shamshad's "saiyaan dil mein aana re" -
> with very different (and loud) orchestration, and sung by a singer
> whose pronunciation was terribly anglicized. Generously sprinkled with
> English phrases all over, and picturized on girls with cyan-colored
> lips, this was a "tampered" product. However, HMV's revival series
> was different from that. Those songs were remastered - recorded again
> in a modern studio with the same instruments as the original, and
> overlaid onto the original voice track. I hope you see the difference
> between these two cases.

I do see the difference between the two cases. In the case of the remix, the song has been re-created by someone. One may not like what she has done with the song, but it is no small mercy that she hasn't used Shamshad's original recording to carry out her excesses. In the case of HMV's revival series, HMV took gross liberties with the *original* versions of the songs. In my book, that is way more egregious than remixing songs.

> Adding color to a gray-scale movie does not change the mood of the
> movie.

Pull the other one - it's got bells on!

Seriously: I can only repeat what I said earlier, that we simply have very different views on the subject. I prefer the original versions, unadulterated. Where there is a need for restoring to address deterioration of the original, I am in favour of strict boundaries that stick to *restoring* and don't try to enhance or modify the original in any way.

(Going by current reports, the work done by the World Cinema Foundation on Uday Shankar's "Kalpana" hews to this minimalist approach to restoring - it will be worth checking it out as and when it becomes available for public viewing.)

Warm regards,
Abhay

[Kalyan]

On Apr 29, 11:42 pm, Abhay Phadnis <aphad...@gmail.com> wrote:
> On Sunday, 29 April 2012 02:35:07 UTC+5:30, AR  wrote:
>
> (snip)
>
>
>
> > Colorization can be considered one form of restoration.
>
> Sorry to respond piecemeal, but I forgot to add that this is incorrect: restoration (especially as it is used in the arts) means bringing something as close as possible to its original form. Adding colour to a black-and-white film, therefore, is not "restoring" it.
>

But the actual scene that was being shot was in colour. So, if that
can be restored with some accuracy (I believe different colours map to
different shades of grey in a B&W film), I consider it a good thing.
Instead of improving the B&W print, which was an approximation, we are
restoring what was in front of the camera.

Regards,
Kalyan

>
> Warm regards,
> Abhay

[AR]

Touche!

Another hidden advantage is that the older movies (Naya Daur, Mughal-e-
Azam) are being *re-released* after colorization. So, it's some sort
of revival. Consequently, the beautiful songs in these movies get re-
marketed as well.

[Chetan]

On Apr 30, 3:56 pm, Kalyan <kkolach...@yahoo.com> wrote:
>
> But the actual scene that was being shot was in colour.

Not if you were to believe Calvin's dad :)

http://multifamilyinvestor.com/wp-content/uploads/2010/03/calvin-hobbes-world-black-white-color.jpg

> So, if that
> can be restored with some accuracy (I believe different colours map to
> different shades of grey in a B&W film)

This is obviously true, but the reverse is not. In other words, the
mapping is not one-to-one. 3 color dimensions (R,G,B) map to 1 gray-
scale dimension. Trying to reconstruct the original colors from this -
even if we assume there is value in it - is like trying to reconstruct
a 3D shape from its projection on a line. It is not possible.

C

[AR]

On May 1, 12:54 am, Chetan <vinc...@gmail.com> wrote:
> On Apr 30, 3:56 pm, Kalyan <kkolach...@yahoo.com> wrote:
>
>
>
> > But the actual scene that was being shot was in colour.
>
> Not if you were to believe Calvin's dad :)
>

> http://multifamilyinvestor.com/wp-content/uploads/2010/03/calvin-hobb...

>
> > So, if that
> > can be restored with some accuracy (I believe different colours map to
> > different shades of grey in a B&W film)
>
> This is obviously true, but the reverse is not. In other words, the
> mapping is not one-to-one. 3 color dimensions (R,G,B) map to 1 gray-
> scale dimension. Trying to reconstruct the original colors from this -
> even if we assume there is value in it - is like trying to reconstruct
> a 3D shape from its projection on a line. It is not possible.
>
> C

Reconstruction of a higher-dimensional entity from a lower-dimensional
one is an ill-posed inverse problem. But it can be solved (to a degree
of approximation) if appropriate constraints are added on the solution
space. These constraints arise from knowledge about what one really
expects to see as the solution. The 3D shape of an object can be
reconstructed from two images of that object taken from very nearby
viewpoints (it's called shape from stereo), making some assumptions
about the 3D shape - such as its smoothness or self-similarity.
Another example is interpolation given a bunch of samples - it can be
done successfully if you make appropriate assumptions about the
function. (Technical digression: under appropriate conditions, it is
possible to reconstruct the entire function *perfectly*, i.e. with
zero error, from a very small set of samples. That's what the field of
compressive sensing is all about).

Similarly, there are algorithms for transferring color onto grayscale
images. I have no idea whether they were employed in the actual
colorization of the black and white movies or what method was really
used. My guess is that perhaps a few color pictures of some scenes
from black and white movies had been already been acquired - and later
carefully preserved. Given a bunch of such color images and their
grayscale counterparts, a mapping function (or a whole set of them)
was "learned". This (set of) mapping function(s) was/were used for
performing the transfer for the other grayscale frames of the movie.
Perhaps, there was quite a degree of manual intervention to correct
for artifacts. The end product looks fabulous from a technical
perspective - i.e. there are no color artifacts or changes in hue.

[Sukesh]

Watch this colorization demo by Ultra.

http://www.youtube.com/watch?v=CzwESL4uSLE

[Sukesh]

Another one

http://www.youtube.com/watch?v=c_fZccD-ocE

[vrk]

do they sell these restored and cleaned versions.

also do they enhance the audio. i am more of a dialogues man, and my biggest complaint with most of the dvds is that i can barely make out the dialogue. the english subtitles rarely help.

[Chetan]

On May 1, 6:10 pm, AR <aji...@gmail.com> wrote:
>
> Reconstruction of a higher-dimensional entity from a lower-dimensional
> one is an ill-posed inverse problem.

Okay

> But it can be solved (to a degree
> of approximation) if appropriate constraints are added on the solution
> space.

My concern is with the "degree of approximation" and the "appropriate
constraints" :)

> These constraints arise from knowledge about what one really
> expects to see as the solution.

This is well and good - not perfect, but perhaps acceptable - for
things like skin tone or the color of trees. But what does one do with
the color of the sky? Or clothes?

> The 3D shape of an object can be
> reconstructed from two images of that object taken from very nearby
> viewpoints (it's called shape from stereo)

Thanks for the short tutorial, but 2 problems. One - I was talking
about a 1D projection as an *analogy* and you are talking about 2D.
Second - we have no 2 images equivalent that I can think of.

> Technical digression: under appropriate conditions, it is
> possible to reconstruct the entire function *perfectly*, i.e. with
> zero error, from a very small set of samples.

Not clear what "samples" are in this situation. Perhaps this is a moot
point for the current discussion.

> The end product looks fabulous from a technical
> perspective - i.e. there are no color artifacts or changes in hue.

Any examples of a fabulous colorization project, preferably from Hindi
Films? Colorization efforts I have seen have been ghastly.

C

[AR]

On May 2, 1:32 am, Chetan <vinc...@gmail.com> wrote:
> On May 1, 6:10 pm, AR <aji...@gmail.com> wrote:
>
>
>
> > Reconstruction of a higher-dimensional entity from a lower-dimensional
> > one is an ill-posed inverse problem.
>
> Okay
>
> > But it can be solved (to a degree
> > of approximation) if appropriate constraints are added on the solution
> > space.
>
> My concern is with the "degree of approximation" and the "appropriate
> constraints" :)
>
> > These constraints arise from knowledge about what one really
> > expects to see as the solution.
>
> This is well and good - not perfect, but perhaps acceptable - for
> things like skin tone or the color of trees. But what does one do with
> the color of the sky? Or clothes?
>

Read the last para of my previous post.

> > The 3D shape of an object can be
> > reconstructed from two images of that object taken from very nearby
> > viewpoints (it's called shape from stereo)
>
> Thanks for the short tutorial, but 2 problems. One - I was talking
> about a 1D projection as an *analogy* and you are talking about 2D.
> Second - we have no 2 images equivalent that I can think of.
>

I was just giving an example (a general one) of solutions to an ill-
posed problem - in this case reconstructing 3D from 2D. I guess I
digressed way too much :-)

> > The end product looks fabulous from a technical
> > perspective - i.e. there are no color artifacts or changes in hue.
>
> Any examples of a fabulous colorization project, preferably from Hindi
> Films?
>

Mughal-e-Azam, Naya Daur and the Chori Chori clips. There may be many
others from English movies.

> C

[Chetan]

On May 3, 8:05 am, AR <aji...@gmail.com> wrote:
>
> Read the last para of my previous post.

I did, before I asked the question :)

> Mughal-e-Azam

ghastly

> Naya Daur

marginally better, but only marginally. Look at the clothes and
jewelry in uDe.n jab jab - not at all convincing

> Chori Chori clips

haven't watched yet

C

[AR]

On May 2, 11:54 pm, Chetan <vinc...@gmail.com> wrote:
> On May 3, 8:05 am, AR <aji...@gmail.com> wrote:
>
>
>
> > Read the last para of my previous post.
>
> I did,  before I asked the question :)
>

If you have learned a mapping from grayscale to color for image frames
of a particular type, then you can use that knowledge to colorize
other frames that are "similar" to the ones you trained on. For
instance, if you had a few color stills of Madhubala in the "pyaar
kiya to darnaa kyaa" song, you can use information from them to
colorize other stills from that song. Of course, the word "similar"
here is vague, but that's where contemporary machine learning stands.
And perhaps, that's why you would need some manual intervention to fix
up any blips.

Again, I have no idea whether they used exactly this method, or if
not, which else. But there are research papers which propose such
methods.

> > Mughal-e-Azam
>
> ghastly
>

I didn't find anything ghastly in the colorizing of Mughal-e-Azam. I
prefer the color version to the black and white one. The next time I
watch Mughal-e-Azam, I won't even think of the black and white
version.

[Chetan]

On May 3, 9:22 am, AR <aji...@gmail.com> wrote:
>
> I didn't find anything ghastly in the colorizing of Mughal-e-Azam.

ghastly vs. fabulous - two poles that can never meet I suppose :)

C

[Kalyan]

On May 3, 9:22 am, AR <aji...@gmail.com> wrote:

Thanks, AR, Sukesh and Chetan, for an informative discussion. With
respect
to defining the constaints, in addition to colour pictures/clips from
the film, I
would assume that people involved in the original film might also
help. But I
would assume that getting this approximation right would involve a lot
of painstaking
manual effort and a lot of to-and-fro/iterations.

The links to the colorization demo by Sukesh (I was hoping they would
show the process)
do make that it clear, at least to me, that the colour version is far
more richer and vibrant,
if done right.

Kalyan