Why/how does newer animation look more vibrant?

[mg262]

While working with some old animation from the 80s, I came to realise that it looked markedly less vibrant than new material and, critically, that none of the software processing I'd tried could change this fact. That included all the standard levels/gamma/brightness/saturation/contrast adjustments, the range of more esoteric things found in the VirtualDub plug-in Colour Mill, and an AVISynth colour adjustment filter (http://forum.doom9.org/showthread.php?t=96308) I wrote myself to make the colour profile of one clip mimic that of another. I then considered the possibility that the effect was caused by a change in texture due to noise, but the screenshots below suggest (at least) that this is not the full story. So... does anyone have any ideas?

Two pictures appear below. The first a still picture from DVD source, which rather usefully happens to persist for two hundred frames or so, allowing very strong temporal denoising to give the second picture.

The key point is that just by looking at either shot you know it's an old cartoon -- or at least, I hope everyone else finds that too. It's very possible that I just haven't found the right filter/setting to use... but you can verify that for yourselves by taking either picture and applying any filter you like in any program... (I will be extremely happy if anyone can prove me wrong here!)

DVD
http://people.pwf.cam.ac.uk/mg262/posts/colour/DVD still.jpg

Denoised DVD
http://people.pwf.cam.ac.uk/mg262/posts/colour/DVD still denoised.jpg

Out of mercy to those on dial-up, those are JPEGs; PNGs can be found here...

DVD (http://people.pwf.cam.ac.uk/mg262/posts/colour/DVD still.png)

Denoised DVD (http://people.pwf.cam.ac.uk/mg262/posts/colour/DVD still denoised.png)

I find it rather hard to put precise terms on the issue at hand... it's not quite fading and it's not loss of saturation; "washed out" is closer;lack of vibrancy is the best I can come up with. I hope it's clearer from the pictures.

One last point, which is that I'm not necessarily looking for filters straight away... this is more an attempt at analysis, and if we can diagnose the cause I'm very happy to try writing filters to reverse the effect. So any thoughts, however speculative, are welcome.

Incidentally, one of the difficulties I have is that while I've seen a certain amount of more modern material, I don't have such things at hand from which to take screenshots... if anyone knows what I'm talking about and has suitable new material, a picture would be appreciated!

[Guest]

LOL.

When you have the "vibrant" one to compare to, give me a ping.

[Koti]

I find it rather hard to put precise terms on the issue at hand...

vhs source

[mg262]

@neuron2...

Do you mean that you think this is purely in my head? Or that it's impossible to diagnose without side-by-side versions?

@Koti,

DVD source. If you mean that you think it was mastered from VHS tape... I think VHS is a consumer format and also if it had been passed through VHS at any stage absence of detail due to the frequency limiting would be apparent.

I have had a request for a VOB sample by PM; I'll put one up in the morning when processor load is low enough to let me produce it.

[Guest]

Do you mean that you think this is purely in my head? Or that it's impossible to diagnose without side-by-side versions? Of course, I don't think you're imagining it. You've made too many expert posts to even think of that. I thought it was amusing that you have this firm idea but no samples of the "good" kind. And yes, it's hard to see your distinction without the good and bad exemplars, not to mention the vagueness of the term "vibrant".

If we had the exemplars, we could do things like calculate color histograms, determine color saturations, image quality metrics, etc.

Maybe the cartoon looks old just because of fashion, e.g., I can tell a sixties movie from the styles of the shirts and haircuts! It's unusual nowadays to see as much texture as you have in the scenery. Art like that takes more time to create and is thus more expensive.

[mg262]

@neuron2,

Of course, I don't think you're imagining it. You've made too many expert posts to even think of that.It is really very kind of you to say that. However I think that while I am reasonably confident with maths and a few technical issues, my intuition about video is markedly less good than that of most of the regulars of this forum -- so I am very open to the possibility that this is in my head. There is also the possibility that, as you say, it is largely a difference in content that creates the effect. I have one snippet of reported information (outlined below) which suggests that there is something above and beyond these effects, but unfortunately it's not something I can verify at present.

Screenshots first. This is a modern example (taken from a clip @MoMonster sent to me for blend analysis):

http://people.pwf.cam.ac.uk/mg262/posts/colour/modern_0.jpg

http://people.pwf.cam.ac.uk/mg262/posts/colour/modern_1.jpg

http://people.pwf.cam.ac.uk/mg262/posts/colour/modern_2.jpg

And to facilitate comparison...
http://people.pwf.cam.ac.uk/mg262/posts/colour/split.jpg

This is clearly not optimal content for comparison! (But is the best I have.) The clearest examples I have seen have been in clips from very new productions that have been posted on Doom9, and unfortunately they seem to disappear rapidly (and I didn't have the foresight to keep copies). The most recent material that I own DVDs of is from the mid-90s, and while there is still a noticeable difference in "vibrancy", or the vividity of the colours, or whatever it should be termed, the effect is far less marked.

I realise I'm not pinning down the effect very well... sorry; it's not for want of effort. My immediate thought was that it was a saturation issue, but increasing the saturation of the original makes parts, especially reds and oranges, look very over saturated. I played very briefly with masks to try and deal with this but to no effect. From the shot I have posted, it looks like there are too few blacks (i.e. that I need to apply levels with output-low at 0 and input-low above zero), but looking at the histogram for the entire clip (which I will post later) shows that this is not appropriate.

Difference in original content does certainly play a part -- for example, I think the dull colour of the clothes on the boy in my original picture exacerbates the issue. There is one other piece of information, which I don't want to attach too much significance to, suggesting that this may not be the entire cause: an old 80s cartoon, the Mysterious Cities of Gold, was remastered from the cels for Japanese broadcast in the 90s; I have read (I think on an associated mailing list) that there is a marked difference in the vibrancy and vividness of the colours of the DVDs made from the Japanese masters compared to those made from the original masters. On the other hand, I haven't had the chance to verify this for myself... The limited-edition Japanese DVDs came in two box sets, selling for $250 each.

I will post histograms, vectorscope pictures, etc. later to save everyone needing to repeat the work. (Also the VOB.) One other thing I have is a copy of the old cartoon that was taped onto VHS in the late 80s, which has aged in a different way... and I have e.g. scatter plots showing how the luma of one relates to the luma of the other (obtained by lining the clips up spatially and temporally). I left this material aside to avoid complicating the original post, but I may put it up later if I can hammer it into a readable form.

Lastly: you have given me one more idea (thank you!)... as you say, the backgrounds differ considerably in complexity, making comparison difficult; but there may be some mileage in concentrating on the (apparently) flat blocks of colour that appear in foreground objects like the dog and boy and analysing those... I will post again one I have some results from that.

[niamh]

Heh, Belle et Sébastien :):)
I have to agree with koti, I think too that it's an original vhs or whatever equivalent ... probably mug funky has all the answers, and we can hope he reads this thread.
Personally I own a few of that type of source, and I never managed to make them look less than poky, if somewhat better than the source.

[Soulhunter]

Dunno, whats with this?

http://img359.imageshack.us/img359/5861/dvdstill9di.png


Bye

[Mug Funky]

old telecines have an amount of colour crosstalk - the CCDs (or whatever the hell they used back then) would pick up a little of each colour when reading the film. also, colour grading until very recently was quite a lossy process - basically placing coloured glass in front of the film as it's copied... this is why new transfers tend to look better (check out old ghost-in-the-shell vs special edition, then compare THAT to GITS SAC, and there'll be more vibrance in each one).

nowadays animation isn't even done on film. it's printed to film if it's meant for a theatrical release, but even that is likely to change soon.

less analog = more vibrant.

you could possibly recover some vibrance with channel-mixer style filters (the most basic being increasing saturation), the advantage being that you can add and subtract channel values to increase saturation just for certain colours, so you can get a tint as well.

i think there's a channel-mixer plugin for avisynth somewhere, and you can practice on stillframes in photoshop. you wont get rid of all the problems of old blurry noisy analog, but you can get a nicer overall result.

[Joe Fenton]

Wow, my art skills finally come in handy. "Old" cartoons/anime were drawn and painted by hand using standard paints. Modern cartoons/anime may be drawn by hand, but they are painted (inked is the industry term) on computers now. Computers are capable of more highly saturated colors than are possible in nature - you can't make physical paints that are as "colorful" as can be generated on the computer. You see this in live action movies as well - for example, the Shire in Lord of the Rings wasn't colorful enough to give the fairytale quality Jackson was looking for, so they had computers process all the frames shot at the hobbit village to make it more vibrant and colorful.

[Mug Funky]

although computers have a hard time reproducing flourescent inks and paints... and anything that absorbs ultra-violet and re-emits visible light. even UV lights show up as blue in RGB.

[mg262]

@Mug Funky,

I knew you would have something for me! :) Thank you very much. Searching suggests that there is a lot of material out there on colour crosstalk and ways of treating it, so I will try to assimilate some of it before writing a proper reply. But please bear with me while I verify one thing: can I read from your answer that as far as you know, there is no effect where the quality of masters degrades over time to produce this kind of effect? (Given the nature of the storage format, it would seem strange to me if there were such an effect, but I don't understand the situation well.)

@Joe Fenton,
Computers are capable of more highly saturated colors than are possible in nature...they had computers process all the frames shot at the hobbit village to make it more vibrant and colorful.Thank you too... that opens the possibility that we may be able to do something similar ourselves, which is good to know.

@Soulhunter (longest last...),

Thank you... that does help to some extent but it seems to have a similar effect to Levels in extending the luma range downwards:

http://people.pwf.cam.ac.uk/mg262/posts/colour/Soulhunter_luma.jpg
(I used your picture as an excuse to try out the early version of ColourLike, which required spatially aligned clips.)

... and that is problematic for the reasons I mentioned above; here are the histograms for the still I posted and for the clip as a whole to clarify that:

Still
http://people.pwf.cam.ac.uk/mg262/posts/colour/DVD still histogram.jpg

Clip
http://people.pwf.cam.ac.uk/mg262/posts/colour/DVD histogram.jpg

I posted that particular still because it was the only one I could denoise in that sledgehammer manner*... but the lack of vibrancy/vividity of colour is apparent even in darker frames (like this one (http://people.pwf.cam.ac.uk/mg262/posts/colour/dark frame.png)), and I think to some degree even in your version. I would like the script though!

*I used TemporalSoften(100,255,255)

@all
I think that to some degree, whatever kind of denoising I tried, part of the effect is due to noise/texture. I've tried marking the regions above various luma values (using the denoised picture) and I hope it brings out the effect:

http://people.pwf.cam.ac.uk/mg262/posts/colour/mark title Sebastien.png
(It seems that some of the noise is spatially structured, and some was added before this frame was duplicated.)

I think I still perceive part of the effect of this as a colour effect (although I can of course see the noise as well). Sometime back I tried replacing animated foreground regions with blocks of solid colour, and the effect just looks wrong. Interesting, but wrong... I think I will post shots to show what I mean, but I have already made everyone read quite enough for one post.

Last thing... although I do really appreciate any improvements that can be made on this video(!), it's the general phenomenon that is worth diagnosing and perhaps trying to reverse... so please don't go to much trouble on these particular samples.

Incidentally, @mods, if this is becoming too specialised for this forum, please do move the thread...

[Soulhunter]

I posted that particular still because it was the only one I could denoise in that sledgehammer manner*... but the lack of vibrancy/vividity of colour is apparent even in darker frames (like this one (http://people.pwf.cam.ac.uk/mg262/posts/colour/dark frame.png)), and I think to some degree even in your version. I would like the script though!

I did it in PhotoShop! But before doing the level tweaking / contrast enhancement, Ive tweaked the colors... Coz, beside the "washed out" look, Ive noticed that this frame looks somehow "nicotized" (maybe they smoke a lot in the film archive? ^^) Well, the dog, the clouds, the text... everything that should be "white" has a yellowish tint. So I raised the blue level a little bit before doing the other stuff...


Bye

[Joe Fenton]

When playing with this stuff in the computer, don't forget the radiosity. This is a measure of how much something glows or produces light. Adding radiosity can really give slick results that aren't possible with just playing with the saturation or brightness.

[Joe Fenton]

beside the "washed out" look, Ive noticed that this frame looks somehow "nicotized" (maybe they smoke a lot in the film archive? ^^) Well, the dog, the clouds, the text... everything that should be "white" has a yellowish tint.

Film degrades over time. The degradation is faster with the cyan and magenta pigments, so the film gradually gets darker and more yellow (remember the three primary subtractive colors are cyan, magenta, and yellow). Look at old photos from when you were a kid (or if you're a kid, from when your parents were kids)... they look distinctly more yellow.

[Guest]

Adding radiosity can really give slick results that aren't possible with just playing with the saturation or brightness. OK, I'll bite. How do you add radiosity?

[mg262]

@Soulhunter,

Yellow: You are completely right. I'd been looking at the (seriously yellowed) VHS version so much that the DVDs had looked slightly blue to my eyes. (I'm sure there is a moral in that... ) In any case, it's very satisfying that your empirical observation perfectly matches @Mug Funky's theoretical one... and it suggests an appropriate countermeasure (gamma-correct to linear light space, try to reverse the now hopefully linear colour mixing to reduce the yellow, and gamma correct back -- although I still have much reading to do on this). Thank you very much!

@Joe Fenton,

neuron2 bit first. I had thought masters were normally kept on tape, not film* but that would also explain the yellow. Plus someone must have modelled the yellowing of film so as to open a reversal methods...
*The one old cartoon for which I checked this listed a UK master of x" tape for some x...

Edit: see below for reasons.

[mg262]

I think the main focus of this thread has been dealt with (and when I am a little more organised I will post a filter I have just written with a friend which deals with texture side of things, which was my particular concern). So I have removed the pictures that were here... but there was one more thing that cropped up and was not dealt with, namely radiosity.
don't forget the radiosity. This is a measure of how much something glows or produces light.I wrote some notes last week relating what I understand radiosity to be on the basis of the above description to the standard methods we use of . In any case, I want to put this up because some months back I spent a while hunting for something similar and had very little luck -- so I hope no one minds the temporary resurrection of the thread; again please move it to the AVISynth forums if it seems more appropriate there. Those who want to skip the maths should look for the heading Conclusion below!

Notes on light

So here is a summary I've extracted from Poynton's colour and gamma FAQs and a few other sources, augmented with a little explanation.

The obvious ways of measuring the amount of light reaching a detector have the following property: if you double the amount of light (by putting two light sources next to each other), then you double the number measured. Measures like this are called linear.

The measures we use here are definitely not linear. I'm going to follow the standard terminology and use the symbols R',G',B' and Y' for the non-linear quantities we normally play with; R,G,B and Y are reserved for linear quantities. Our standard measures are obtained from the linear quantities like this:

R' = R^gamma
G' = G^gamma
B' = B^gamma

The number gamma depends on what standard you are using, but for Rec. 709 it is 0.45. (In fact a different formula is used for very low values but I will leave that aside for now.)

Y' is then obtained as a weighted average of the non-linear quantities R', G', B'. In Rec. 601. the formula is

Y'=0.299R'+0.587G'+0.114B'

So, just emphasise it again, all the quantities we use are marked with prime signs (') above. The quantity Y' which we use in e.g. YV12 is called luma. (Luminance is something different. Incidentally the Y in the XYZ colour format is luminance, not luma.)

So back to the original question... how do we boost light in a linear way (something that would correspond to doubling all the light sources, if you like)? Well, first we go from the Y'U'V' colour spaces in which we *usually* work (namely YUY2 and YV12) to R',G',B', which is something we do all the time. Then we have the reverse the above transformations to get the linear light components:

R = R'^(1/gamma)
G = G'^(1/gamma)
B = B'^(1/gamma)

Then we e.g. double these (or multiply them by whatever factor k we like),...

Rnew = k R
Gnew = k G
Bnew = k B

...and then go back to R',G',B' using the first set of equations, and probably then back to Y'U'V'. All of this should probably be done with higher than 8-bit accuracy...

For the particular case that we are playing with, as opposed to any other linear operations, the equations simplify to:

R'new = k^gamma R'
G'new = k^gamma G'
B'new = k^gamma B'

And because R',G',B' can be mapped to Y', U', V' by a matrix, it follows that

Y'new = k^gamma Y'
U'new = k^gamma U'
V'new = k^gamma V'

Conclusion
So in this case by increasing radiosity we're just applying a levels operation to the luma channel and a saturation boost to the chroma channel.

If we were doing something more complicated to the linear channels R,G,B then it would not have been possible to translate the operation to anything straightforward in terms of our usual channels -- in particular, if there had been channel mixing of the linear channels, to reverse the effect we would have to apply a gamma transformation, reverse the mixing with an appropriate matrix, and apply another gamma transformation.[I]

I am very prone to making silly errors, so please read through the above with a critical eye...

[Joe Fenton]

OK, I'll bite. How do you add radiosity?

Most people do it when they're making the original artwork since it's far easier. Remember that most objects are just reflecting light, subtracting certain percentages of certain colors to give an overall color. Radiosity works by making the object a SOURCE of light as well as a reflector.

This is easy to do in 3D rendering programs, which many anime artists use quite a bit these days. I think Photoshop has some ability to generate luminous colors in 2D. I know of several commercial packages which are designed to make existing art more radiant. I used one on our company logo years back for some CD covers we needed.

[Mug Funky]

hang on... is there really any way to add diffuse reflections to a 2d image? i suppose it could be spoofed in photoshop if one were to spend a lot of time on it, but thats more in the realm of creation than restoration. the maths are astronomical for a 3d scene, let alone a still 2d image (i'd wager it's impossible, but nay-sayers typically get proven wrong).

looking at the VHS image, the washed out look isn't just lack of saturation (it's actually quite colourful), but more from the glows around bright objects - possibly introduced during telecine, or maybe when the original print was made. also, outlines suffer a fair bit from general blurring. the same picture with darkened lines would probably look a lot "newer", but the glow would have to be tackled as well, or it'd just look weird.

film masters do degrade over time, but the colours stay separated, as they are different layers on the film. the film will just drift toward red over time, and this can be fixed pretty easily with a per-channel level correction.

[Joe Fenton]

There's a fine line between art restoration and art re-creation. That's why art galleries pay big bucks to talented restorers. If you've ever watched the science channels and seen a show on restoring old masterpieces, you'd see that they are literally repainting the artwork in many cases as nothing remains of the original.

This is true of film as well. Watch the Star Wars extras about restoring the original trilogy - especially the original film. It was ALMOST beyond repair, and many talented artists were employed to rework the film into something presentable.

Making the anime at the top of the thread look "new" would most likely mean virtually redoing the whole thing from scratch with the original serving almost as the storyboard.

[mg262]

looking at the VHS image, the washed out look isn't just lack of saturation (it's actually quite colourful), but more from the glows around bright objects - possibly introduced during telecine, or maybe when the original print was made. also, outlines suffer a fair bit from general blurring. the same picture with darkened lines would probably look a lot "newer", but the glow would have to be tackled as well, or it'd just look weird.Oh, now, that's really interesting... I had noticed the effect before, but I hadn't realised it mattered:

http://people.pwf.cam.ac.uk/mg262/posts/separation/frame 329.jpg http://people.pwf.cam.ac.uk/mg262/posts/separation/frame 329 background.jpg

... the brightening around the white dog is very clearly visible. (As you say, I had tried line darkening, but it just looked bad. And line sharpening is something I've never really seen made to look good.) But that kind of halo effect is something that should certainly be counterable, at least to some degree -- for example by lowpass filtering and subtracting a multiple of that from the original.

By the way, the picture is from a DVD (I have VHS tapes of the same and the difference in detail level is visible) -- do you mean that it originally came from a generic tape source?

Edit: I just came back to check that the glow happened in all directions, implying that it is not anything to do with a tape -- of course, you'd got there already:glows around bright objects - possibly introduced during telecine, or maybe when the original print was made

Edit: So that kind of thing works pretty well to my eyes (original on the left, filtered on the right):
http://people.pwf.cam.ac.uk/mg262/posts/colour/antiglow_371.jpg

http://people.pwf.cam.ac.uk/mg262/posts/colour/antiglow_32967.jpg

That was calibrated to leave blocks of flat colour alone, to reduce the levels issue, though it is still there to some degree. I've also adjusted both before and after from levels range 16-235 to 0-255, which is something I probably should have done before ...