Need help removing VHS noise (cartoon video question)

[Colmino]

I have recently acquired DVDR encodes of some old cartoons which were on VHS tapes. The bitrate is fine and MPEG2 artifacts are reasonably scarce.

The trouble is the VHS artifacts. They're pretty bad. There seem to be several types, which probably have real terminology but I can only provide descriptions. If anyone knows what these are really called, feel free to chime in.

1: Grainy noise. This looks to me sort of like enhanced film noise.

2: Ghosts. This looks sort of like the dreaded edge-enhancement you frequently get in DVD movies, only it isn't 360 degrees. It seems to be two or three "echoes" that can be seen to the immediate right of the main image. Particularly visible on strong image contrasts, meaning in this case the black contours surrounding every character, since these are cartoons.

3: Sync problem at the bottom. A familiar sight, to say the least. For whatever reason, the bottom 2% or so of the screen seems to always suffer some sort of corruption when it comes to VHS video. As far as I can tell, there is no way to recover the video lost to this corruption.

The obvious first question is: Are there any standardized filters or procedures which are known to provide some sort of results as far as removing these sorts of artifacts?

My less obvious second question: Since these are cartoons, and they very definitely do not play at 30fps (more like 6 to 12), it seems to me that if one were so inclined, and tenacious enough for the job, one could take several sequential fields of video (or, rather, frames generated from the fields) which correspond to the same exact frame of cartoon animation, and somehow extrapolate a new "idealized" frame of animation from the information garnered from the less-than-ideal sequential frames. For example, let's say frames 2000 to 2005 happen to be the same exact frame of cartoon animation. Would there be a program out there that is designed to (or capable of) sort of "merging" these six frames into one? This would produce a rather more pristine frame of cartoon animation, which could then be substituted in place of the original frames 2000 to 2005, with the result being superior video, at least for those six frames.

Any help appreciated!

[redwudz]

I am most familiar with VirtualDub Mod for filtering. Many filters available at http://neuron2.net/. Unfortunately, this would require re-encoding to AVI format, which will result in further loss. I don't use AVIsynth that much, but from others, it may be a better option for applying filters.

I would at least experiment with VD or AVIsynth and see what you can do. Not a lot of MPEG filters out there.

And this post seems entirely related to restoration, moving you to that forum.

[guns1inger]

neuron2.net is definately the place to start.

Fileters worth looking at include

msmooth and msharpen : both created specifically for processing cell animation (mainly anime). msharpen comes in both vdub and avs flavours. msmooth is avs only.

exorcist : designed to deal with ghosting. The type of ghosting you described sounds like poor reception, rather than over sharpening. Exorcist is vdub only.

[Colmino]

Unfortunately, this would require re-encoding to AVI format, which will result in further loss.

Well, I assume it will have the option of producing raw video. I don't have a problem with large filesizes, as the video segments I intend to process are not lengthy.

The type of ghosting you described sounds like poor reception, rather than over sharpening.

The ghosting I poorly described is not related to reception as in broadcast reflection because this video was never broadcast. I likened it to edge enhancement because the ghosts were immediately next to the original image, in a diminishing ripple pattern very similar to edge enhancement. I've seen it many times, and readily associate it with nth-generation VHS video. The ghosting is possibly picked up in inadequately-shielded wiring between the VCRs. But yeah, it's certainly not actually edge enhancement. Just very similar, and definitely not classic broadcast ghosting.

I'll be interested to see if anyone can offer advice on the idea of combining frames to form one "merged" frame. This would be tremendously enabling.
[/quote]

[guns1inger]

exorcist still sounds like the right tool. There was an avisynth version called ghostbuster in development, but it seems to have died.

Can you describe more fully your idea on merging frames ? I'm not quite sure what you are getting at there, but it does sound interesting.

[lordsmurf]

1- Noise removal filters in VirtualDub (static noise reduction, dynamic noise reduction) or TMPGEnc, both are fine.

2- Not a lot can be done. Play with filters in VirtualDub, but do not expect much

3- Are you sure this is not just overscan? It will not be 20%, but closer to maybe the lowest 3% of the image at most, maybe a few on the top and sides as well. It's not VHS as much as it can be from the source too

Cartoons are done 20-24fps or so and often edited and finalized native to 29.97 (25fps PAL gets converted from the 29.97 version). So unless these are cartoons from 100 years are, your assumption is incorrect.

[Colmino]

Can you describe more fully your idea on merging frames ? I'm not quite sure what you are getting at there, but it does sound interesting.

Cel animation almost always hovers at a framerate lower than the actual framerate of the media it gets played on, be it film or video. This is perfectly understandable. Production costs for hand-drawing 24 or 30 (or 60) cels per second would be prohibitive. This means of course that at any given moment in a cartoon on video, many fields are devoted to what is technically the same image. For example, fields 2000 to 2005 might correspond to the cartoon's animation frame number 250. Of course this is not to say that the cels dictate the framerate of the animation (scrolling may occur at higher framerates, for example), or that the cel animation is going to remain constant (usually it varies). But in the interest of providing detail on this particular idea, we will assume that such considerations are irrelevant.

In this example, I would isolate fields 2000 to 2005, generate deinterlaced frames from those fields, and then merge all six frames together to create a single "averaged" frame. There would be every reason to anticipate that this "averaged" frame would turn out to be closer in quality to the actual pre-VHS iteration of the animation frame than any of the original six VHS fields managed to be.

Note: I do not imply that there might possibly be an app out there that can perform this dynamically for an entire program. Obviously it would need to be done on a handcrafted level, cel by cel.

2- Not a lot can be done. Play with filters in VirtualDub, but do not expect much

That's pretty much what I expected. I figured there had to be at least some program or filter that one could use to reduce the phenomenon ("Exorcist", in this case) but wasn't looking for miracles. Even though there's no reason to doubt that adequate processing could eventually come very close to correcting the artifact.

3- Are you sure this is not just overscan? It will not be 20%, but closer to maybe the lowest 3% of the image at most, maybe a few on the top and sides as well. It's not VHS as much as it can be from the source too

You're talking about the same thing. I suggested 2% (not 20%) but I wasn't certain of the percentage. I associate it with VHS and other tape media, but mostly VHS since it seems to be a problem that can be pinned on poor-quality (cheap consumer) tape-recording electronics.

While I don't really expect there to be a single solution to this common problem, I figured it's common enough that maybe somebody took a stab at figuring out some way to partly fix the corruption. After all, even though the affected portion is corrupt, it was nonetheless clearly part of the original image, and not some random noise (such as closed-captioning lines).

Cartoons are done 20-24fps or so and often edited and finalized native to 29.97 (25fps PAL gets converted from the 29.97 version). So unless these are cartoons from 100 years are, your assumption is incorrect.

I believe my explanation was misinterpreted. I comprehend the matter of media framerates. I was speaking of the effective framerate of the cels themselves, which of course is only very rarely 24fps, such as in Disney features (and then not all the time).

[Colmino]

Can you describe more fully your idea on merging frames ? I'm not quite sure what you are getting at there, but it does sound interesting.

Subsequently, I have located this webpage where they describe a hardware device which seems to be designed to perform pretty much the precise function I have attempted to describe, albiet dynamically on a constant video source.

http://www.colorado-video.com/noisereduction.html

Here is a relevant portion of text from that page:
" The Colorado Video Model 449 Dynamic Video Noise Reduction System can filter or reduce dynamic video noise (like "snow" or "static") out of a monochrome video signal by essentially averaging frames over time. This type of video noise filtering or reduction is best suited for signals with rapidly changing noise in pictures of relatively static (unchanging) subjects. "

So it's not a new idea. Obviously I'm looking for a software solution rather than what I presume to be an exceedingly expensive piece of hardware, and nothing fancy or dynamic. How difficult could it possibly be to take six frames or images and generate an average of them? There pretty much has to be something out there that can do it.

[lordsmurf]

You can cover (mask) the overscan up in the encoder or editor, but honestly, who cares? You'll never see it on the tv, and better LCD/plasma system have overscan filters. I would do it just to save bitrate if doing AVI captures (with latter MPEG encode after editing), but if you capture direct to MPEG-2, forget about it. I guess since you'll be restoring this, and in an encoder and editor, go for it.

I don't know where you're getting information on hand-drawn frames only be 6 or so per second, but it's wrong. Generally, they hand draw at least 12-15 frames, sometimes up to 18-20. But that style died a decade ago (throughout the 1990s, for the majority of them), after being around since the 1930s-1950s or so. Yes, it was a lot of labor, but your main animator was the only person to draw 6 or so frames, and secondary animators did the latter in-betweens, and then others yet again came along with ink and colors. These days, you draw the main frames, and then computers act as the in-betweens. More these days, sometimes even a full 30 frames in anticipation of HD tech, with ability to add 60fps if they wanted to. So what you ask cannot be, as there are more frames. If you drop below 15 frames, there is clear movement distortion.

I don't know if you've ever been to an animation museum (Chuck Jones has or had a nice one) or if you own any animation frames.

[Colmino]

You can cover (mask) the overscan up in the encoder or editor, but honestly, who cares? You'll never see it on the tv, and better LCD/plasma system have overscan filters.

Masking the corrupt video portion is probably the route I will go. Or possibly I will look into scaling the image. For you see, the method I use to watch most of the video material I have is by use of an HTPC. As one might anticipate - and it is by strong preference - the full resolution of the image, including whatever was intended or at least produced as overscan, is shown on the screen by the various media playing applications. I differentiate between intent and non-intent because most of the time, the overscan video is no less valuable or pertinent than any portion of a movie that gets chopped off in a pan&scan process, and is therefore subject to the same justifiable preference.

I don't know where you're getting information on hand-drawn frames only be 6 or so per second, but it's wrong.

Not to belabor a point that's rather off-topic, but I can only say that you and I don't watch the same animation. Most animation (and by that I might clarify to mean Japanese animation, since it represents the bulk) does indeed hover at framerates between 6 and 12. In fact, since most such animation is frequently still-shots with animated mouths, one could reasonably argue that the framerate really averages much lower.

As for it being information that I'm "getting" from somewhere, nonsense. I watch the material myself, and I have an adequately firm grasp of the visual representations of varying framerates.

Back on topic, though, it seems that a piece of software was developed which takes the concept of merging frames to the next level, way beyond the humble solution I seek. It's called Visar, and you can watch a brief program about it here:

http://www.exn.ca/news/video/exn2003/04/28/exn20030428-visar.asx

Naturally enough, given how advanced the software's features are, it's not even a consumer product yet.

I'll be seeing if perhaps Photoshop has the ability to combine X images into a single image.

[lordsmurf]

I don't know where you're getting information on hand-drawn frames only be 6 or so per second, but it's wrong.

Not to belabor a point that's rather off-topic, but I can only say that you and I don't watch the same animation. Most animation (and by that I might clarify to mean Japanese animation, since it represents the bulk) does indeed hover at framerates between 6 and 12. In fact, since most such animation is frequently still-shots with animated mouths, one could reasonably argue that the framerate really averages much lower.

Japanese animation is far more advanced (not better, mind you, just more advanced), since the late 1970s and early 1980s (Gundam and Macross, for example), than your average American or European work. The 1980s work was very well done hand-drawn work (Thundercats, for example, from Pacific Coast Animation), and 1990s and 2000s work is completely done with computer assistance (Robotech Shadow Chronicles, for example, from Tatsunoko Productions).

All I can say is that what you think is happening is not the reality of the situation.

Also do not confuse it with frame layering (bodies do not move while mouths do, what you refer to, which rarely happens) and backgrounds. That "bodies do not move while mouths do" sort of motion-comic-book style is something I've really only seen in the 1966 Marvel cartoons (Mighty Thor, Captain America, and a couple others).

Maybe this is not true of the exact show you refer to, you have something obscure or homebrew, but for a general discussion on how animation works, you're far off base.

There are 2-3 times the frames involved than what you're stating (minimum of 12), and then those are doubled yet again for the final video. In the old days, that meant doubling up some frames, but more modern times means you just have a computer program invent more in-betweens for you, gives more fluid motion. That can also help if you wanted to create pure PAL from pre-rendered sources, but I don't see much of that, all the PAL stuff I see is half-assed NTSC conversions, for the most part. I've seen some pretty cruddy PAL releases (check out Transformers Robots In Disguise from UK, that release is poorly done).

I disagree this info is off-topic. You suggested this was relevant to a restoration method, and that is just not the case. You would probably damage motion/fluidity to a great degree with that sort of method.

Your "Visar" link does not open an asx video. I can only assume you're referring to NASA's VISAR technology used in forensic video restoration, and I do not think it would work for this sort of application.

[Colmino]

When I have developed a slightly more specific approach to my questions about VHS-source cel animation restoration, I will generate a more on-topic thread. In the meantime, since my first attempt is already squarely hijacked, let's proceed.

Japanese animation is far more advanced (not better, mind you, just more advanced), since the late 1970s and early 1980s (Gundam and Macross, for example), than your average American or European work.

I won't argue. There are varying reasons, but it could quite reasonably be suggested that the primary one is the fact of differing social attitudes towards, and therefore differing philosophies in the creation of animation. I'll limit my comments on this tangent to that, as I never really sought to discuss animation philosophy.

All I can say is that what you think is happening is not the reality of the situation.

I respectfully remain adamantly confident. More likely my descriptions and your comprehension of said descriptions are meeting with incompatibility.

Also do not confuse it with frame layering (bodies do not move while mouths do

Always good to add to the list of terminology selections.

what you refer to, which rarely happens)

I really have to respond to this. Can you be serious? I have two terabytes of Japanese animation which suggest otherwise. The so-called "mouth flapping" phenomenon is a staple of anime, right up to and including the most recent and popular TV productions like Bleach, Naruto and One Piece. The most common formula is for people to move a bit, freeze, mouth shares three separate frames of animation which are cycled semi-randomly, repeat. It is likenable to the Hanna-Barbera school "head bobbing" staple, observable in just about every single HB cartoon from the 60s and 70s, where animation is chiefly limited to the occasional up/down motion of the speakers' heads. And it serves the same purposes, although in the case of anime it's likely also a stylistic choice, since it can be observed even in Miyazaki features.

There are 2-3 times the frames involved than what you're stating (minimum of 12), and then those are doubled yet again for the final video.

Once again I believe you are failing to see which "framerate" I am referring to. I would say it was my failure but I think I've done an adequately thorough explanation by now. But just to reiterate: I refer to the framerate of the hand-drawn cels. Visualize for a moment that the animator has the task of animating, say, a woman standing in place while her hair flows in the wind. The animator has decided that he will cycle six individually drawn cels of this woman in such a fashion that the flowing hair appears perpetually animated, without interruption. He has also decided that the duration of this cycle will be 1.0 seconds, to correspond to the target media's 24 frames per second, which translates to six cel frames per second, and also means that each of those cels will be exposed to four frames of film. There's really nothing else to it. But to make everyone happy, let's take it for granted that this film eventually winds up on a PAL DVD, making it easy and painless to secure a full-frame digital rendition.

And of course the frame-merging concept I wish to undertake comes into play here. As mentioned before, each cel gets four whole frames all to itself. But there's noise. In my case, VHS noise. But certainly also minor mpeg2 artifacts and even ultimately some film noise. So you combine those four frames into one. Voila. A single frame whose quality (assuming the image is adequately still between each frame, which it is) is most probably superior to any of the four original frames, and which now can be substituted in place of the four original frames.

Now take this train of thought a small step further. One has a VHS recording of some animation which fits this bill. What they're primarily aiming to reduce is VHS noise. Fortuitously, VHS provides a lot of frames to work with, meaning, in the case of this animation, a lot of redundancy. Means the likelihood of results is good.

As can be observed in the Visar link I provided earlier, this isn't just some improbable crackpot concept I pulled from under my hat. It's far from original. It's been done.

I disagree this info is off-topic. You suggested this was relevant to a restoration method, and that is just not the case. You would probably damage motion/fluidity to a great degree with that sort of method.

It might prove helpful to give my posts more than a cursory glance, as I stipulated well in advance that I was not seeking a process which would enable me to restore entire programs in one go. If I am aiming to reduce noise on select sequences, and there are no mitigating considerations (such as scrolling or other phenomena happening at framerates which differ from that of the cels), then it will work just fine.
[/i]

[lordsmurf]

to correspond to the target media's 24 frames per second, which translates to six cel frames per second, and also means that each of those cels will be exposed to four frames of film.

And this is the part that is not correct.

Until you realize how animation works, with key frames and in-between frames (which is virtually EVERY non-key frame in the digital age, because of computer generated cels), you're not going to get very far with what you want to do. All you'll do is butcher the hell out of your video.

There are more than 6 animated cels in a 24 fps (or 25 or 30 or whatever) video. Most of this stuff is not shot to film anymore either, hasn't been in ages. In fact, a lot of this stuff is now even drawn directly into a computer with a tablet.

Anyway, good luck to you ..... it's clear you don't want to listen to somebody who has experience in restoring video (with a specialization in animation methods).

[Colmino]

And this is the part that is not correct.

With respect, it's spot on. Not only is it starkly intuitable, but I have already isolated a few "theoretical" sequential frames of film / video from a few different anime and, sure enough, the individual frames are effectively identical, differences in noise artifacts notwithstanding.

Until you realize how animation works, with key frames and in-between frames (which is virtually EVERY non-key frame in the digital age, because of computer generated cels), you're not going to get very far with what you want to do. All you'll do is butcher the hell out of your video.

Where you became confused from the onset is the fact that I wasn't looking for some sort of legitimizing nod from bonafide professionals as to the feasibility of my proposal; I understood perfectly well that it would work. All I was after was the name or names of the software tools which would enable the process. Photoshop may be a solution. I do not yet know.

Anyway, good luck to you ..... it's clear you don't want to listen to somebody who has experience in restoring video (with a specialization in animation methods).

Sir, to be perfectly honest, given the studious manner in which you circumnavigated the facts I painstakingly laid before you, I had already come to the conclusion that this had become less an effort to give advice and more one of assertation of superior knowhow. Well, congratulations, end of thread, moving on.

[Marvingj]

The most commonly used method today to compress digital video data is MPEG-2. From current satellite streams and digital cable feeds to off-the-air digital broadcasts, MPEG-2 has now been internationally adopted for a variety of applications.

MPEG-2 first exploits temporal redundancy through motion estimation and then proceeds to spatially subdivide the image in 8x8 blocks upon which the DCT (Discrete Cosine Transform) is applied to exploit spatial redundancy. Compression is done by quantizing resulting DCT coefficients and re-ordering them to maximize the probability of long runs of zeroes, and then run-length coded. Finally a Huffman encoding scheme is used. The whole process allows for great savings in terms of bit-rate ratio (>10:1).

However, these savings don’t come free, and because the codec discards some of the original video information, there can be serious side-effects; MPEG-2 is what we call a lossy codec. It discards image information believed to be of lesser visual importance. The more you want to compress, the further away you get from the look of the original image. Image quality and fidelity now depends on the chosen (or often imposed) level of compression. And since that is directly tied to the available bandwidth, we must ask ourselves when is the video simply too compressed?