Color Correcting the Dragon Box - 3 Part Spectacular

[ect5150]

I'm time constrained today, so I'll have to read the whole thing later but I have a few questions.

Is it possible to perform the color cast removal step via AVISynth instead of using VirtualDub?

Is it worth it to some to start a project somewhere else to make these files and keep them for those that want to correct our DBox colors as a group and share?

[Son Geeko]

I'm going to go out on a limb and ask: Is this possible with the orange bricks?

I know a lot of peopkle will ask what the point is, but I'm simply asking if the orange bricks are still in a workable state, and whether a POC exists.

If someone did this and actually succeeded, I would give that guy $25.

[Metalwario64]

I'm going to go out on a limb and ask: Is this possible with the orange bricks?

I know a lot of peopkle will ask what the point is, but I'm simply asking if the orange bricks are still in a workable state, and whether a POC exists.

If someone did this and actually succeeded, I would give that guy $25.

You mean like, making them look better overall, or just simply altering the color palette and nothing more?

If the latter, then of course you can. It's not as if Ashura's methods make Photoshop, Virtualdub and the other programs suddenly only compatible with the Dragon Boxes.

If the former, then well, I've still got some old attempts at making season two look "passable" lying on my hard drive...:

Season Set:


Edited:


Season Set:


Edited:


Ignore the cropping (lol) of the sides, and the bars on the sides of the last image. Those were other things I was screwing around with on those images back when I made'em. The PQ is the point of them though. Unfortunately, for episodes like this:



There's not much you can do, as they were so bright that entire details were just wiped out entirely.

This was the best I could do:



Sometimes the skin is crushed white though, so you have shots like this:


That can only be made to look like this:


Or if you want to add some red for skin tone like this:


Looks like a weird Gohan Kulilin hybrid... Just goes to show how bad season two can look.

Maybe you'd settle somewhere in between with white skin, but a bit darker and more bearable to look at with at least the shading "recovered"?:


Do I win $25, or do I have to make this work with Ashura's method?

[BlazingFiddlesticks]

Edited:


Omigosh, this actually looks like a decent single shot!

[bkev]

You know I wonder if the Orange Bricks did in fact get the blu-ray treatment (let's hope not...) if there would be more image that could be recovered by tampering with the colors. Star Wars fans have managed to take out much of the crushed blacks, though your personal taste on their choice of color pallet will vary - as with Dragonball, for that matter.

[BlazingFiddlesticks]

They are native 1080p prints, after all.

[Corpsecreate]

Excellent guide Ashura. The pictures made me understand this a hell of a lot easier (I'm a photoshop noob) than the private message you sent me a while back. The process is actually quite quick once you know what your doing and I managed to load the virtualdub plugins via avisynth (after uninstalling the hacked MT avisynth.dll) which is great.



I have a question I have been trying to figure out, maybe you know the answer to it. Do the dragon boxes have a colour range of 16-235 or 0-255? Since Dragon Ball was always intended for older CRT TV's it would make sense to me that the range would be 16-235, however I am not sure.

The other question I have is about the colour cast. If there is a green colour cast, does that mean that the film has more green in the print than intended or does it mean the Red/Blue has diminished while the green hasnt? If its the former, then it would make sense to pull down the whites to the lowest value of the RGB when adjusting the curve, if its the latter then you would have to pull them up to the highest as you described.

[Attitudefan]

Reminds me of a poorly kept beta tape or VHS of the Saban dub! I was thinking of doing this a while ago but never got around to it.

[Corpsecreate]

I dont see why you would want to colour correct the orange bricks when you can use dragon box footage?

[Zestanor]

I have a question I have been trying to figure out, maybe you know the answer to it. Do the dragon boxes have a colour range of 16-235 or 0-255? Since Dragon Ball was always intended for older CRT TV's it would make sense to me that the range would be 16-235, however I am not sure.

Correctly made DVDs only have a brightness range (Y) of 16-235. Values 0-15 and 236-255 would be ignored on almost everything (CRTs included.) 16 is the darkest black, and 235 is the brightest white. IIRC, the buffer of unused values was designed so that overshoot interfered less with the actual picture, when recording digital video, and this carried over to DVD, too. Anyway, on an LCD or Plasma monitor, The 16-235 is stretched to the native 0-255. So if there was any info with 0-15 or 236-255, it would be lost. On a CRT, 16 is set at 7.5 IRE, which is basically zero output (actually, in Japan 16 is set to literally 0 IRE, because Japan's CRTs could handle that, or something.) CRTs don't really have a brightness limit, I mean you could keep making it brighter until it exploded, I guess. Still, 235 would be at the hypothetical brightest point.

Make sure not to relate the 0 and 16 difference with the 0 IRE and 7.5 IRE difference. These are not related.

[Attitudefan]

I dont see why you would want to colour correct the orange bricks when you can use dragon box footage?

To see if it was fixable in the least.

[Ashura]

The other question I have is about the colour cast. If there is a green colour cast, does that mean that the film has more green in the print than intended or does it mean the Red/Blue has diminished while the green hasnt?

It means it has more green in the print than it should, since it's found in the white areas as well. (It sort of seems like the film has 'yellowed' ... but instead 'greened'.) If it was the other way -- where it /just/ lost blue and red data -- I don't think the eyes and other 'white' objects would be affected in the way it is here.

BTW, all the terms in this article are my own. They are probably not any kind of industry standard.

[Corpsecreate]

Correct me if I am wrong but the greyscale are colours that are composed of equal amounts of Red Green and Blue. This would mean that if only the red/blue diminished then there would be a higher concentration of green thus turning the whites into a 'white with a shade of green'. The same would also be true if the red and blue stayed the same and the green was increased, the only difference is you would get a lighter shade of green. Heres an example of what I'm saying:



So either way, you get more green in the image, the question is which one is it. If the colour range is indeed 16-235 then that would mean that if there is any scene with a value higher than 235, then it must mean more of that colour has been added to the film. I have taken a picture of a white screen from episode 4 and I got (233,255,235) as my RGB. This should mean that there is more green than there should be and so when doing the white balance, the curve should be taken down to the lowest number instead of up to the highest (not the best explanation of what I'm trying to say but you should understand).

I ask this purely out of interest, I'm not trying to start any sort of argument or anything. I'm merely pursuing a path that hopefully leads to more knowledge

[ect5150]

If the colour range is indeed 16-235 then that would mean that if there is any scene with a value higher than 235, then it must mean more of that colour has been added to the film.

This doesn't sound right. I'm not a color space expert, so others should chime in here... but the color space that DVDs use is apparently limited to that range (16-235) for brightness. The color space isn't an RGB one for DVDs. While it's more natural for us to think in terms of RGB, it doesn't mean that's the best thing to encode for compression efficiency. DVDs use YUV I believe... there are simple equations (which i never bothered to learn) to alter the values from RGB to YUV.

Therefore to get something "white" the YUV brightness value (the Y in YUV) must be maximized... which is a value of 235. The other two values tell it there is no color (versus a brightness maximized shade of blue). It's more efficient to compress colors vs. brightness than it is to compress three separate color channels.

[ect5150]

Anyone know the relevant line to import the vcf file settings into an AVISynth script? I prefer AVISynth and I'm not have much luck finding the info via Google.

EDIT: Nevermind. Got it to work. It's a pain in the ass, not a simple one line item.

[ect5150]

Question 1

Ok... so for the color cast removal, what is the general recommendation or something like this:


Click to Enlarge

There is a grey-ish mouth area, but the RGB values are roughly R=216, G=200, B=204. So, it's slightly red heavy, but that bright white area has some pixels balanced on 253,253,253. This comes from one of the "green sky" episodes, so we know the colors are off. Does this basically imply the color cast isn't there to remove?

EDIT:
Question 2

With the curve generated in the transform filter (trying to remove the color cast), isn't this most likely going to lead to the following effect?

Click to Enlarge

Certain color values above 230 (or whatever number is on each color channel), shouldn't this result in some of the "bleeding edges" most people didn't like the Orange Boxes? What's the best way to alter the colors without this effect seeping in?

[Zestanor]

That right there is a beautiful shot. I love what you guys are doing, but when I saw that cap, I remembered why I love the Dragon Boxes. That is an example of graceful fading. That's why we were tricked for so many years that the DBoxes had the original colors. That is a scan of the 16mm film that came out of the camera that photographed it. From what I've seen, any kind of artificial color correction to it comes off as, well, artificial. Even in this thread I've seen comparisons, and the Dragon Boxes just look more natural; easier on the eyes. Sure, the colors may be way off, but in my opinion, the original colors for this episode are 20 years long gone. Newsflash: the cells aren't correct either! Funi's are closer, yes, their film was copied from a copy that was printed before these faded, but then that makes them multi-generational. They have extra layers of grain, and are not as clear as these. The colors on theirs are also faded, but in the sense that all colors have lost some saturation. They look stale and flat, while the O-negs' colors are pleasing and vibrant, even if the actual values are technically incorrect.

Still, please carry on! Like I said, I love what you're doing, even if given the choice, I'd rather watch the pristine 0-gens in all their wacky color shifted glory.

EDIT: My post was in response to the Kuririn image.

Also, remember: In most image/video analyzing software, it will display absolute black as 0, and white as 255, but that's only because it stretches it out from the 16-235 stored on DVD.
In fact, forget about 16-235 if it is confusing you. It's only a problem when you get to the encoding side of video.

[Ashura]

There is a grey-ish mouth area, but the RGB values are roughly R=216, G=200, B=204. So, it's slightly red heavy, but that bright white area has some pixels balanced on 253,253,253. This comes from one of the "green sky" episodes, so we know the colors are off. Does this basically imply the color cast isn't there to remove?

Since we're trying to middle the greys, I would go for the mouth area instead of the bright white spot. We know those are supposed to be the grey values. There are going to be portions of the color that are completely white even if it still has color cast.

Certain color values above 230 (or whatever number is on each color channel), shouldn't this result in some of the "bleeding edges" most people didn't like the Orange Boxes? What's the best way to alter the colors without this effect seeping in?

I've found that this may happen sometimes, but it's mostly an affect of the mpeg compression it seems like. We're working with already compressed materials, so unfortunately we have to deal with problems like this. The highly compressed sky of the shot you posted may be a problem, for instance.

[jpdbzrulz4sure]

I did another YouTube video of color-corrected shots, this time applying color curving and saturation. Just click "attempt #2" in my signature and let me know what you think.

EDIT: It turns out there was some minor artifacting on Goku's neck in the shot with Goku and Piccolo in profile, so I went back in and touched it up.



Though, the dislikes outweighing the likes makes me wonder if there's still something I'm doing wrong.

[ect5150]

Certain color values above 230 (or whatever number is on each color channel), shouldn't this result in some of the "bleeding edges" most people didn't like the Orange Boxes? What's the best way to alter the colors without this effect seeping in?

I've found that this may happen sometimes, but it's mostly an affect of the mpeg compression it seems like. We're working with already compressed materials, so unfortunately we have to deal with problems like this. The highly compressed sky of the shot you posted may be a problem, for instance.

What I'm getting at isn't really a compression issue. Let me try to be more clear.

Take this non-DBZ screenshot:

(click to enlarge)

Now assume we go through the same process as you mention in the guide. We could wind up with something similar in the transform:

(click to enlarge)

Notice my transform filter (Note: I intentionally increased the RGB color channels uniformly to make the explanation a bit more clear... hopefully)
Notice in the red sun, the previous bright spots now "bleed white." This is similar to some Orange Brick screenshots (Bulma bathroom scene possibly?) where the "white bleeding" is basically destroying some of the detail. Now, the cause is obvious. The transform filter is greater than 45-degrees (or has a slope greater than +1, whatever). This means previous values of red are now pushed up too high. For example, a pixel with a red value of 215 is now pushed up to 255 at a max (and the same thing applies for green and blue depending on the curves). This will always happen as long as the slope of that line is greater than +1.

So, my real question is this: Can we do anything about it? Is it possible to still remove the color cast without increasing the slope of the line past +1. Instead of adding red where needed (which is causing this effect). Is it possible to just reverse the other color channels instead (remove the green and blue)? This way we should be able to avoid the "white bleeding" as I call it, but still return to the greys to "proper greys." I'm thinking this may darken the picture though, since we are removing color instead, but it might be worth a shot?

Also - for everyone trying this out - try to tackle DBZ episode 34. This is one of the "green sky episodes". I'm having a very difficult time getting the sky to be a nice blue without causing artifacts or have Vegeta's uniform turn purple. I would love to see some others try to tackle this episode and see what the results are.