Strange artifact looks like interlace but apparently not

[xuguang_he]

It seems that the red striped vest the girl wear has some stranged artifact.
The first shot is strange, but the second one return to normal. These two shots were taken within a few seconds.
I've got no idea how to deal with it.


Screenshot was taken from the attach video clip which gave a more straightforward sight.
Wishing someone could help me out.

[sanlyn]

The m2v is progressive with 3:2 pulldown (telecine). Inverse telecine might make the stripes look smoother, but basically it's a digital artifact known as moire. It's a common occurrence in digital video. There are ways of smoothing it somewhat, but no one that I know of has devised a truly effective way of eliminating it.

[Brad]

I would say it's not moiré and instead point to the MPEG-2 encoder as the culprit. If you step through the video at PAR you will see smooth wave-like distortions.

Animated PNG:

[Attachment 17764 - Click to enlarge]

[jagabo]

The video was interlaced (telecined) at some point in the past and whet through a bad IVTC to make it progressive.

[Brad]

How do you figure? The waves, sail, etc. don't suffer along diagonals. Not to mention this is clearly from a HD master.

What is the film, anyway?

[sanlyn]

I see those scratchy ripples frequently, even on retail DVD's. I'm sure there must be a dozen causes, including encoders, DVD players, TV's, and whatnot. I've never seen anything that will clean it up.

[poisondeathray]

I see those scratchy ripples frequently, even on retail DVD's. I'm sure there must be a dozen causes, including encoders, DVD players, TV's, and whatnot. I've never seen anything that will clean it up.

Anyways - There is aliasing on the boat hull as well, and it occurs on the exact same frames that the "moire" patterns occur on the bikini top => this supports the comments the defects are caused by an error in processing rather than optics or the sensor during acquisition . Moire patterns during acquisition will occur in different areas and patterns and not in this distribution

Regardless of the cause, using a general antialiasing filter will help a bit, but damage other parts of the picture (small details will be erroded) , unless used selectively through masks

This type of repair is typically done with motion tracking , clean plates ,with something like mocha pro , after effects . You can think of it as taking areas from surrounding "good" frames and moving , rotating, scaling them like a "patch" to "cover" over defect areas . It sounds difficult, but most of the "heavy lifting" is done by motion tracking . It's semi-automated with the right tools, but there is manual work involved and it has to be done by scene. You would use a similar approach on the boat hull or any other areas with problems

[sanlyn]

I figured there was a way, but not a simple one. Will save this little clip and try it in After Effects later.

[jagabo]

How do you figure? The waves, sail, etc. don't suffer along diagonals.

Yes they do.

Not to mention this is clearly from a HD master.

1080i to 480p.

[Brad]

Frame 0:
[Attachment 17771 - Click to enlarge]
Frame 0 bob:
[Attachment 17772 - Click to enlarge]
Frame 0 TDeint:
[Attachment 17770 - Click to enlarge]
Frame 1:
[Attachment 17773 - Click to enlarge]

Show me an example of poorly deinterlaced 1080i being visible in a 480p downscale.

[vhelp]

@ op, it would help if you gave us more details about the source clip "snap.demuxed.m2v" origins. how was this clip obtained.. is this clip an extract of one of your encodes from elsewhere.. did you capture this clip from analog or digital source, etc. etc.

otherwise, because this artifact seems more confined to that area, i would suspect the artifacts are a bi-product of some boched ivtc / filter combo on your part. or, if it came from some hd source, (receiver perhaps) then that device has a bad motion deinterlace/inverse telecine downscaler.

[xuguang_he]

Oops, so there's no way to solve it in Avisynth?

BTW, the movie is Love Has Many Faces (1965), and this is not a downscale source. It has been released through Made-On-Demand DVD only, never has HD appearance. If it has a bad motion deinterlace/inverse telecine downscaler, I won't know it since I'm not a personnel of Sony Home Entainment.

[Brad]

It's available to own in HDX on Vudu for $14, so there is an HD master and this DVD certainly wasn't sourced from an old telecine.

[jagabo]

Since there are some good frames here and there You can try using ReplaceFramesMC() with a mask.

https://forum.videohelp.com/threads/352741-Frame-interpolation?p=2226119&viewfull=1#post2226119

The mask was just a fuzzy white box marking the area with the bikini top:

Mpeg2Source("snap.demuxed.d2v")#, CPU=2, Info=3) 
Trim(0,97)
src=last
mask=ImageSource("mask.png").ConvertToYV12(matrix="PC.601")

ReplaceFramesMC(1,1)
ReplaceFramesMC(3,7)
ReplaceFramesMC(16,22)
ReplaceFramesMC(39,5)
ReplaceFramesMC(45,19)
ReplaceFramesMC(68,26)

Overlay(src,last,0,0,mask)

I stopped at frame 97 because there wasn't a good frame at the end of the video. But the woman moved so little you copy probably just copy frame 97 to fill it out. The technique didn't work well for frames 3 to 10 but the rest looks pretty good.

[poisondeathray]

Is whole movie like that? or only this scene ?

The timing of the "good" frames doesn't make sense. If anything it confuses what might have happened

point to the MPEG-2 encoder as the culprit.

I'm curious why you think that ? I've never seen an encoder do this

[xuguang_he]

Only these few seconds. Other parts although interlaced, could be decimate with: tfm(order=1).tdecimate()

[xuguang_he]

Since there are some good frames here and there You can try using ReplaceFramesMC() with a mask.

https://forum.videohelp.com/threads/352741-Frame-interpolation?p=2226119&viewfull=1#post2226119

The mask was just a fuzzy white box marking the area with the bikini top:


[Attachment 17779 - Click to enlarge]

Mpeg2Source("snap.demuxed.d2v")#, CPU=2, Info=3) 
Trim(0,97)
src=last
mask=ImageSource("mask.png").ConvertToYV12(matrix="PC.601")

ReplaceFramesMC(1,1)
ReplaceFramesMC(3,7)
ReplaceFramesMC(16,22)
ReplaceFramesMC(39,5)
ReplaceFramesMC(45,19)
ReplaceFramesMC(68,26)

Overlay(src,last,0,0,mask)

I stopped at frame 97 because there wasn't a good frame at the end of the video. But the woman moved so little you copy probably just copy frame 97 to fill it out. The technique didn't work well for frames 2 to 10 but the rest looks pretty good.

Not bad solution, but just take a look at the edge of the boat, bad aliasing still appear here and there.

[xuguang_he]

It's available to own in HDX on Vudu for $14, so there is an HD master and this DVD certainly wasn't sourced from an old telecine.

To my knowledge, they won't play tricks on this DVD. The HD source maybe just transformed from this SD source.

[jagabo]

Not bad solution, but just take a look at the edge of the boat, bad aliasing still appear here and there.

Of course. I didn't address that.

[manono]

Other parts although interlaced, could be decimate with: tfm(order=1).tdecimate()

Wait a minute. Are you saying the whole thing was hard telecined? And you gave us the result of your IVTC? Because there's no interlacing in the sample. The sample is soft telecine - encoded as progressive with 3:2 pulldown. If what you gave us is the result of your IVTC and your reencode, I think we have the culprit - you.

Or have I misunderstood what you said? Or is it maybe a mix of hard and soft telecine?

[xuguang_he]

Other parts although interlaced, could be decimate with: tfm(order=1).tdecimate()

Wait a minute. Are you saying the whole thing was hard telecined? And you gave us the result of your IVTC? Because there's no interlacing in the sample. The sample is soft telecine - encoded as progressive with 3:2 pulldown. If what you gave us is the result of your IVTC and your reencode, I think we have the culprit - you.

Or have I misunderstood what you said? Or is it maybe a mix of hard and soft telecine?

Of course not, the clip I gave you is the original DVD clip, I just try the TIVTC on avs.
Also, the strange things appear on the girl's bikini is apparently not interlacing, it's not merely a coincidence I should say, many other sources I've come across with also have this kind of "wavy".

[Brad]

I'm curious why you think that ? I've never seen an encoder do this

The artifacts appear to me to be blurred waves and not what I think of as a moiré effect (though I guess this is similar). Why would random single frames in the middle of the sequence turn out okay?

To my knowledge, they won't play tricks on this DVD. The HD source maybe just transformed from this SD source.

Vudu doesn't accept SD sources to sell as HD.

[poisondeathray]

I'm curious why you think that ? I've never seen an encoder do this

The artifacts appear to me to be blurred waves and not what I think of as a moiré effect (though I guess this is similar). Why would random single frames in the middle of the sequence turn out okay?

Moire looks exactly like this in motion when the camera moves. Obviously this is a film scan, but moire from digital sensors can occur at certain angles and depths and yet be perfectly fine when you move the camera just a few cm . If you've ever shot progressive video with DSLR's they line skip because the sensor is so large, you get these same patterns.

To me, this looks more like an error in processing because basically everything is affected in the "bad" frames, not just a certain focal plane. But with an error in processing you would expect every frames to be affected (maybe some slightly less if there was a cadence pattern) but there is no pattern here . The fact that only this scene is messed up also adds support that someone just messed up.

So I don't know why there are "good" frames at all - I basically asked the same thing above. The pattern doesn't make sense. But at least you can make use of them as clean plates

Only these few seconds. Other parts although interlaced, could be decimate with: tfm(order=1).tdecimate()

Since it's only a few seconds I can do a quick fix it for you. Here is the motion track/patch repair method done on the side front of the hull, and the bikini top .

The aliasing in the center part of the boat would take more time than I'm willing to spend (you would have to roto out the man, the woman, shadows, moving ropes, it would take probabaly over an hour to get it perfect) , so I just used antialiasing filters with rough masks in that part

(FFV1, lossless YV12)

[fvisagie]

This type of repair is typically done with motion tracking , clean plates ,with something like mocha pro , after effects.

Just for my own education, how is http://en.wikipedia.org/wiki/Match_moving related to the motion tracking you mention? Where can I read more about motion tracking? And what are clean plates? From Google it seems to have a lot to do with healthy eating.

Thanks!
Francois

[jagabo]

Motion tracking: http://www.videocopilot.net/basic/tutorials/05.Motiontracking/

[poisondeathray]

This type of repair is typically done with motion tracking , clean plates ,with something like mocha pro , after effects.

Just for my own education, how is http://en.wikipedia.org/wiki/Match_moving related to the motion tracking you mention? Where can I read more about motion tracking? And what are clean plates? From Google it seems to have a lot to do with healthy eating.

Add "VFX" to "Clean Plates" in your search term...

In general , a "clean plate" is a "good" frame - so no noise, errors, or unwanted objects (even actors) . In VFX work, a clean plate is often a background shot without the actors or foreground objects. It's used so you can digitally "subtract" objects or "add" or composite things into the scene as if they were originally there. eg. Think of movie stunts where you have actors suspended on wires - you want to remove the wires . It's easy to do if you have the exact same shot (identical camera move) with the background, without the wires . Or, similarly if you want to add CG elements like robots, spaceships, aliens , anything, to the scene - if you have clean plate sequences the work is much more trivial

Clean plates go hand in hand with motion tracking and match moving. "match moving" is more specific and usually refers to 3D tracking; "motion tracking" is more generic term and can mean a 2D motion track only (a flat plane, only x, y coordinates) or can include match moving. For example if you want to add makeup or scars to actors' faces, maybe remove blemishes , and you want the effect to "stick" to the face, you need to use motion tracking techniques . If you want to add a 3d model, some dinosaur to the scene etc... you need to match move

In match moving - If you shoot a live action sequence and want to place something into the scene , you need to reconstruct the geometry of the scene as if the 2D elements captured in the film/video existed in a 3D space - objects farther away in the horizon would have different x,y,z coordinates than something up close to the camera. So both the camera movement, and objects in the scene movement must be tracked. So the 2D objects from the film now have a 3D representation

[fvisagie]

Thanks, guys!

Cheers,
Francois