Restoring new life to old film classics

[carlmart]

I have two old John Ford color films that I would like to put new life into: "Drums along the Mohawk" and "The quiet man".

I have the DVDs that apparently were mastered from poor quality originals, and I don't seem to find new/better masterings anywhere. They do not seem to attract enough attention to reconvert the original films to HD or anything like that.

So I am willing to try a recovery of my own. In order to do that I wonder which avisynth filters should I try.

My idea is to write a script and then playing with them on-line in AvsP.

Suggestions?

[G)-(OST]

Hey

Did you know there was a dedicated restoration forum here, you might want to ask a mod to move this topic over there so people can help you better.

[jman98]

I will ask the mods to move this for carlmart.

[carlmart]

Thanks, folks. That would be great!

[jagabo]

What filters you use will depend on what's wrong with the video.

[jman98]

"Drums Along The Mohawk" was re-issued in region 1 in December 2007. See here:
http://www.amazon.com/Drums-Along-Mowhawk-Ford-Collection/dp/B000WMA6GY/ref=sr_1_4?ie=...8886774&sr=8-4

I don't know if this is different from what you have or not.

"The Quiet Man" was reissued in May 2007 as part of a John Wayne set, but one review said the picture quality was worse than VHS.

[carlmart]

I would very much like this topic to be moved to Restoration. How can I get that?

[minidv2dvd]

as the original poster you might try changing the title to include the word "restoration". the people interested will most likely find it no matter what forum group it's in.

[freebird73717]

There is a thread on doom9 that discuses using avisynth for 8mm film restoration.
Mind you this was aimed at 8mm restoration so you would probably have to adjust some settings for dvd restoration. But at least it could give you a starting point.

http://forum.doom9.org/showthread.php?p=1297663#post1297663

I haven't even tried it but it might be worth a look to you. He posts the exact script he uses and has even posted a video showing the results side by side.

http://www.vimeo.com/2823934

[carlmart]

as the original poster you might try changing the title to include the word "restoration". the people interested will most likely find it no matter what forum group it's in.

Went for a verb instead of subject. Would that make people come in hordes?

[carlmart]

There is a thread on doom9 that discuses using avisynth for 8mm film restoration.
Mind you this was aimed at 8mm restoration so you would probably have to adjust some settings for dvd restoration. But at least it could give you a starting point.

I'm quite familiar with Videofred's 8mm film restoration. He uses a smart avisynth + virtualdub combo to get his results. I am helping a friend restore a 16mm film video conversion with serious problems, and Fred's suggestions showed interesting results.

I just thought I might find a different approach here. Maybe some magic formula to get things back easily.

[freebird73717]

Gotcha.
Hope you get replies from people more knowledgeable than me. Restoration is not my forte.

[carlmart]

For a start what i did was doing some trimmed tests with and without <lsfmod(smode=5)>. Then made DVD-RW of the tests, using different film excerpts. The principle was that a little resolution increase might help.

Well, it did not.

At least viewed on my plasma screen the uncorrected scenes look more natural.

The problems I see to correct anyway are color bleeding (a video problem common to VHS too) and something I would call "noise", but you may call something else and be welcome, in things like grass, flowers or leaves. Probably some detail boost that was applied on the original video.

Let's hope this description is enough and I won't need to upload video parts for you to see what I mean.

[2Bdecided]

Colour bleed?

mergechroma(aWarpSharp(depth=20.0, thresh=0.75, blurlevel=2, cm=1))

from here...
http://forum.doom9.org/showthread.php?t=132066

Cheers,
David.

[jagabo]

Chroma blur also usually results in the colors being shifted to the right (analog low pass filter --> phase delay) so a ChromaShift to the left is usually in order.

[carlmart]

Colour bleed?

mergechroma(aWarpSharp(depth=20.0, thresh=0.75, blurlevel=2, cm=1))

from here...
http://forum.doom9.org/showthread.php?t=132066

Where's the filter on that topic?

[carlmart]

Have a look at the bleeding I am talking about. Watch the woman's nose and the white cloth below her face.

[jagabo]

ChromaShift(c=-2)

[carlmart]

It looks better, Jagabo.

But I am getting the following error message:

Input to filter must be in YUY2 or RGB32 - use ConvertToYUY2 or ConvertToRGB32.

Is it the same to use either?

[jagabo]

I converted to YV12 first (had to adjust the frame size of the JPG image too, YV12 requires mod4). You should get slightly better results working with YV12 from the MPEG 2 decoder.

[vhelp]

@ carlmart

the error you are describing is more than like the result of a decoding issue due to your un-optimized import--assuming you used the avenue of avisynth scripting, however, though I could not see evidense of your issue from the pics that you and jagabo posted so far. But please be aware that if you are trying to post evidense of troubled sources, it would be wise to *not* use jpg, instead, you should use PNG since it is lossless and compresses very small though not as small as jpeg. Anyway.

What is the origin of your source(s): ??

A) dvd recorded to -r/+r media, or
B) purchased commercial dvd

..ripped to HDD, scripted and imported into a video timeline.

Most commercial dvd's that are film based are true 24p (progressive) content and will not show (or result in) these decoding errors (unless you did something wrong in your scripting)

The error is usually because of how the telecined (non-progressive) content get decoded--they (interlaced/telecine) are decoded as progressive (the incorrect way) and in these cases (now that you know this) when scripting these telecine sources that you include in one of the conversion functions, interlaced=true, so that the error is eliminated.

i.e., converttoyuy2( interlaced=true ) should fix the problem in most cases--test first because sometimes the video does not need this after all. Test with and without this paramater. Do this just after the MPEG2source() function.

Still, its possible that another user-error (script or import to editor timeline) caused this, like incorrectly resizing, etc. as jagabo hinted earlier about.

Also, it is almost rare to resort to applying "color-bleeding" type fixes to commercial dvd content though there are cases of such situations--you should be able to review such sources first to determine if they are in facts these erorrs or some other, and then know which correction type application to appy.

As for VHS contents (as capture to an avi container) these problems among others are usually cause by multi-level processes. Such as capture card or equipment (ie dvd recorder) or encoder and decoder during capturing, and of course, the usual importing into a video timeline with its various causing problems as well. All these and more can cause issues on multiple levels, sometimes tricky or deceaving to determine and fix. Sometimes a simple change in capture card will fix the problem(s) or decoder, and so on and so forth.

Same for laserdisc content (just in case you're planning on including some projects in this area) they have their share of issues too, though mostly dot-crawl and a common misconception: s-video vs. composite, where you should always Input/Output to composite connections: laserdisc[composite] -> [composite]capturecard, taking note of the above and so on and so forth.

-vhelp 5161

[jagabo]

I could not see evidense of your issue from the pics that you and jagabo posted so far.

Note how the green hue from the background has spilled over into the woman's nose (carlmart's jpg on the left, 4x nearest neighbor enlargement):

And note how shifting the chroma channels 2 pixels to the left has fixed the problem (on the right).

[carlmart]

the error you are describing is more than like the result of a decoding issue due to your un-optimized import--assuming you used the avenue of avisynth scripting,

Which of the two errors I described: the color bleed or the flickering on leaves and flowers?

Sorry if my photos were not too clear, but I can't seem to capture images as it should be.

The original source is a PAL commercial DVD. The only thing I did was demuxing the original files in order to add smaller subtitles and remaster it, which is something I'm doing with all my DVDs in order to adapt them to new format screens. Many films I have were letterboxed 4:3 types, so I am blowing them up to full 16:9. In all cases I am modifying the subtitles to use smaller letters and longer character lines. Being yellow instead of white, and smaller, they do not disturb so much when you watch the movie on a larer 16:9 screen.

The errors I mentioned do seem to be due to conversion problems, as the video is PAL, but probably coming from the time it was telecined. Only a fresh modern film/video conversion would completely eliminate these problems. But as I am doing this subtitle job on both films, why not try to see what can be done?

[jagabo]

Note that my ChromaShift(c=-2) was just a quick hack. And All I really looked at was the woman's nose. It may be that the U and V channels need different amounts of shift and they may also need vertical shifts. I would look for sharp vertical and horizontal colored edges somewhere else in the video to figure out exactly how much adjustment is necessary.

Also, with YV12 encoding (as is used on all DVDs) you will always get some color blurring because the chroma channels are encoded at half (each axis) resolution. For example a 720x480 video has the grayscale (luma) information encoded at 720x480 but the color information is encoded at 360x240.

[carlmart]

Note that my ChromaShift(c=-2) was just a quick hack. And All I really looked at was the woman's nose. It may be that the U and V channels need different amounts of shift and they may also need vertical shifts. I would look for sharp vertical and horizontal colored edges somewhere else in the video to figure out exactly how much adjustment is necessary.

I looked at the filter page to see how to exactly handling this filter. Certainly because of my still avisynth crawling learning stage, I couldn't quite understand how to manage the variables. Can you explain them a bit to me? What are U and V exactly?

Apparently there are 7 variables: C, U, V, L, R, G, B. You just get them right by trial and error?

I noticed two or three shift moments: one that with the woman I showed (vertical?) and some more involving sky vs trees or sky vs wagon (horizontal?). Let's suppose I correct them: will that affectly wrongly other parts of the image? Or it's just the opposite: things will get better all over?

One thing I foresee: I will need a much better monitor than the one I have now to better find/correct the shifts.

Also, with YV12 encoding (as is used on all DVDs) you will always get some color blurring because the chroma channels are encoded at half (each axis) resolution. For example a 720x480 video has the grayscale (luma) information encoded at 720x480 but the color information is encoded at 360x240.

There's not much I can do there, can I?

[jagabo]

What are U and V exactly?

Basically, U and V are the color information of the video. The Y in YUV is the grayscale brightness. Colors are created by subtracting U and V from the gray.

http://en.wikipedia.org/wiki/YUV

Usually if colors are shifted both U and V will have shifted by equal amounts. And the entire frame will have shifted by the same amount, even the entire video (unless different parts of the video come from different sources). C in ChromaShift is used to shift both U and V horizontally. L is used to shift both U and V vertically.

I recommend you just experiment. Shift by significant amounts so the changes are obvious. And use a video that is more colorful.

[carlmart]

OK. As I am willing to learn a bit (or a lot) more on this "restoration" thing, I am uploading a piece from another Ford film: "The wings of eagles". This film also needs some improving, at least the copy I have.

From what I have seen on other versions you can get here from classic movies, like several Ford's, swashbuckling
and westerns, they come not from prime rate originals. But I am not willing to go after more expensive versions from abroad, as I can use this chance to improve my abilities to get things better with avisynth tools.

So here we are: I trimmed one piece where I can see several things that could be corrected. Hopefully you will show me some more to improve the general look.

The file is here:

http://www.mediafire.com/file/nfmyqzjnjkz/wings_A.m2v

1) Color shift. Now that I learned to look for it, that was easy. If you look at the two larger size actor scenes, you may see some color bleed on the left, on the face and ear. That looked much better with ChromaShift(c=-2).

2) The sky looks "dirty". Probably had film grain on the original master and the compression created some artifacts there. That I don't know if can be corrected. Suggestions?

3) Halos. There are several that might need correction, but I don't know if they can.

4) Sharpening. It looks as if the image was sharpened and created several of the problems I described, particularly 2 and 3. In fact I did try LSFmod and things got worst. Maybe some softening is in order? But that may make things look out of focus, won't it?

What else can you see that I can not that might be improved?

Comments would be welcome.

[jagabo]

I didn't see much chroma shift in that clip.

The video is indeed over sharpened. That is the cause for the halos and some moire artifacts. Blur(1.0) will get rid of most of the halos but leaves it a bit fuzzy. You might try less blurring or following it with LSFMod() to sharpen it back up, hopefully without adding the halos back.

Try using UnDot() to remove some of the DCT ringing artifacts.

The video is pretty noisy. Noise doesn't bother me much but you can clean that up with Neat Video pretty well. You will lose a little of the smallest, low contrast, details -- like the spray in water behind the plane.

[carlmart]

I didn't see much chroma shift in that clip.

What I am starting to see is that getting to a good result is a sum of several things, each one working just a little on a certain end. I did try the chroma shift correction and results improved a bit.

The video is indeed over sharpened. That is the cause for the halos and some moire artifacts. Blur(1.0) will get rid of most of the halos but leaves it a bit fuzzy. You might try less blurring or following it with LSFMod() to sharpen it back up, hopefully without adding the halos back.

I am having some difficulty in finding the blur filter. It's not listed as such in the Filter Collection or anywhere else. Where do I get it?

Try using UnDot() to remove some of the DCT ringing artifacts.

This I am trying as we speak. Burning a test to see how it looks on the plasma.

The video is pretty noisy. Noise doesn't bother me much but you can clean that up with Neat Video pretty well. You will lose a little of the smallest, low contrast, details -- like the spray in water behind the plane.

Noise does not bother me either. What I don't like is artifacts that happen when noise happens.

This is what I am using till now

ChromaShift(c=-2)
ConvertToYV12
UnDot()
spline36resize(720,4800
Trim(8000, 9000)

[manono]

I am having some difficulty in finding the blur filter. It's not listed as such in the Filter Collection or anywhere else. Where do I get it?

It's built in. Just use it without loading anything:

Blur(1.0)

I think you can use any decimal up to and including 1.54.

Personally, when removing halos I prefer to use a halo-specific filter, one such as BlindDeHalo:

http://forum.doom9.org/showthread.php?t=74003

or one of the several others:

http://avisynth.org/mediawiki/External_filters#Dehaloing