I was taking a look at the releases of Peter Pan again today. It struck me that while the Special Edition DVD is slightly better than most other versions, it is far from perfect.

Here are some SBS screenshots showing a simple dehalo.

I also tested adjusting the white balance since this transfer skews blue, but everything I tried made things look more yellow, not more white. Maybe someone has a suggestion for that.

Without trying to match to any particular previous release, and just hitting the black & white points, is this more like it? (open image in new tab for full size) . .




Note that the umbrella was used for black (R=G=B) and the eye (not the shirt, which is not white) was used for white (R=G=B). For slightly stronger red in the skin-tone, I bumped up it's contrast value (rather than use another function's saturation). Yellow highlighting denotes my input values. Also, I found I had to recheck and re-tweak the black & white points after other adjustments (gamma and contrast), so keep that synergy in mind.

BTW, what did the image look like before DeHalo-ing it, and which version / what settings did you use to getting it looking so nice?

It's actually split screen. The left side is the original image, the right is Didée's DeHalo_alpha script with default settings: https://forum.doom9.org/showthread.php?p...post738264

I have no idea if those colors are correct, but they do look more natural. Is that photoshop? I think I did something similar with Gimp once and then translated those numbers to an avisynth Tweak() setting.

Edit: Just looked up the thread where I talked about what I did... and it's not the same thing. I used Gimp to find peak value and used those numbers with Tint().

I wondered if it was split screen or just titling (I'm big on style). 

So zooming in really close, I see the difference between DeHalo_alpha before and after. The original halo-ing also has JPEG artifacts (?) that may be from the DVD or/and from the .JPG save. (I always save to .PNG so as not to introduce additional artifacting.)

I'm using an old wanna-be Photoshop-style paint program (Jasc Paint Shop Pro). It works well with only a few glitches, which I know and maneuver around.    The Histogram function (with it's individual R-G-B settings) is used to test my approaches to fixes. If I need special curvatures, I use the Curves function (R-G-B again) with an eyedropper for precise curve-point settings. (Avisynth doesn't have a one-for-one process to translate my single-frame tests to it's video processing ... I think.) I try to keep away from any other functions as I don't know exactly what they are doing to the spectrum, but will use them for quick testing of a picture.

And speaking of which, I like stronger color that I think was always a Disney trademark, especially in all the "classic" animations. Therefore I tried pushing up the R contrast to increase only the red saturation (and having to retweak the high & low points to black & white again). It's probably not the right way to increase it's saturation, but I haven't delved into that part of spectrum manipulation, yet. Here's the results of that test (open in a new tab for full size) . .



The first set is your original captures, the second and third sets are my previous G-B settings with the updated R adjustments. The most saturated (as seen in the hair color) is my favorite setting.

The skin tones look so much better too.
Now to figure a way to accurately choose some colors and to get it into the film.

As far as artifacts, the original had a lot of compression artifacts/mosquito noise. I didn't want to start cleaning that up until I was serious about considering a restoration of sorts.

Being concerned that those couple of samples didn't represent the full spectrum of color for my test correction, I looked up an old posting I once saw (on another forum) for a bunch of sample samples. Of course, your avatar-names-sake was it's OP   and I found the original link to the releases comparison posted there @ Mediafire - Peter_Pan_v2.7.html.

Here I've applied the exact fix with the "maximum" (-50) contrast on the Red separation, for inspection (open image in a new tab for full size):



Overall, it's starting to look like a Disney film again! But, in a few frames, I see that the Green separation is a little too strong at the low end. Perhaps it should have some contrast, too, then some reduced gamma -- to lessen the low end while keeping the rest of the spectrum the same.

I see what you mean about the green. The second Tinker Bell shot has her dress starting to blend into her legs like the Diamond Edition.

On the whole, very pleasing though.

Edit: Wow, I just noticed that your tweaked colors look awfully close to the 1990/1 laserdisc now. The LD might be a bit too yellow, but it's otherwise much closer to that than the Platinum and Diamond colors.

(2017-11-22, 04:56 AM)Doctor M Wrote: ... your tweaked colors look awfully close to the 1990/1 laserdisc ...

And I wasn't even trying for that, just the Black-White balance and the color of the skin and hair. Maybe the 1990 LD is the real deal for coloring after all! 

.

(2017-11-21, 11:03 PM)BronzeTitan Wrote: Perhaps it [green] should have some contrast, too, then some reduced gamma -- to lessen the low end while keeping the rest of the spectrum the same.

Well, the contrast/gamma didn't work out -- the green curve needs tighter manipulation than Histogram can give it. Some programs include low-middle-high adjustments. That would make it easy just to reduce the Green low-end without touching the rest of the spectrum. Since this paint program doesn't have that, I would then use the Curves function to follow the previous procedure and then reduce the Green's lower area to suit.

I'll be honest, most of what you're doing is over my head. I like what I'm seeing and wouldn't mind working on this or letting someone else.

I think once you hit the Platinum and Diamond editions, too much damage has been done.

Can you make any suggestions on what the next step is? What could be a guide for the proper colors and how to actually get those colors into AVISynth or something besides a photo editing program?

I use the Histogram manipulate as a proof-of-concept guide. Depending on how that goes, one could seond-guess what they might've done to put the film into it's present state. If successful, that can be used as a guide for Curves to fine tune it, like with the too-strong low-end green.

Using that approach for Curves, I first checked for areas of crush and blown primaries (RGB). These pictures had some of that from Red being pushed down for their bluish result. The first thing I did was to lock-down the picture to the 16-235 range. In this case, the lowest value of each R-G-B was moved up to 16 (shown as "0,16). The highest value was moved down to 235 in each, too (shown as "255,235). Note that I use the minimum number of points to approximate the Histogram's curve, just o guide the curve. Trying to "micro-manage" it will generate unwanted artifact-colors in odd places.

Then I looked around to see what needed tweaking (like the strong, low-end green) and other wrong coloring, of which I found. Normally, that's where I stop. But here, I wanted to expand a little of the near bottom of the spectrum to show some of the detail in the shadows that had been pushed together. (Since my monitor is not calibrated, I don't know if this last setting was a valid adjustment. It seems to look a little better on my end.) I didn't want to adjust each R-G-B individually, so I used the RGB-curve, which applies one setting equally to all. Here are those settings . .


. . and here is the resulting comparison (open in a new tab for full-sized) . .


I miss the loss of the super-green Peter Pan costume. Oh, well, that's the trade-offs. So, correction is do-able. Probably the real next step is getting a hold of some original source(s) to guide what colors should be.

If Avisynth fails to have a filter to handle the settings data (I've looked and can't find one that just takes the numbers and runs), there is a really good VirtualDub filter that can do it (and it's interactive, too!). I'll look for where and then come back with a link.