Artifacts In Progressive SD Animation Source

[LouieChuckyMerry]

Happy Saturday! and thanks in advance for any help . After learning (and forgetting, ha ha) quite a bit in the process of shriveling my Simpsons NTSC DVD's, I'm attacking my Futurama NTSC DVD's and had a couple questions. I figured that Futurama would be a breeze because the source is progressive with pulldown flags instead of interlaced but, while the quality is indeed much improved, there are still annoying artifacts. Here's a small sample:

S1.E1-FuturamaTestClip

My questions:

1) What is this called, combing?

2) Why does this exist in a progressive source?

3) How would I best eliminate this using AviSynth, the TFM portion of TIVTC?

Thanks for your time.

[KarMa]

Well there is certainly telecine happening, instead of interlacing. So a field matcher like TFM would correct that. Normally there would be a standard pattern of 3 progressive frames followed by two combed frames, that could be field matched to 4 progressive frames follwed by an extra duplicate. Then using tdecimate() to remove that extra 1 in 5 duplicate frames. But your sample does not seem to follow a simple pattern, as I see it has a long string of none duplicates at the end of the sample (the part with the old man moving his arms)

LoadPlugin(".............\TIVTC.dll")
tfm(order=-1)

You can remove the combing with this. I don't know about returning the frame rate from 29.97 to 23.976, since the pattern is not simple.

[jagabo]

Cartoons are rarely easy to deal with.

DGIndex shows the clip is only 95% film -- so 5% is encoded interlaced. As pointed out by KarMa, you'll need to field match, then decimate to 23.976 fps. But there's something else odd about the pulldown that requires a second pass through TFM. You should use DGIndex in Honor Pulldown Flags mode to build an index then:

Mpeg2Source("S1.E1-FuturamaTestClip.d2v") 
AssumeTFF()
TFM(chroma=false, pp=0) 
AssumeBFF()
TFM(chroma=false) 
TDecimate()

The first TFM uses pp=0 so the post processor doesn't deinterlace frames where the field matching failed. That allows the second TFM to properly field match the remaining interlaced frames. That got rid of all the combing in the luma channel but there's still some minor interlaced chroma artifacts (watch Fry's mouth). Hopefully it won't show up in large areas in other parts of the episode.

And beware that cartoons are often a mix of frame rates so decimating to 23.976 fps may not be right for the entire episode.

[Brad]

https://forum.videohelp.com/threads/373405-What-s-the-best-way-to-de-interlace?p=240434...=1#post2404348

[LouieChuckyMerry]

Thank you all for your answers, I appreciate them and understand them (for the most part, I think). What I don't understand is why once the progressive source, at 23.976 fps, is striped from the pulldown flags that it still needs to be decimated; isn't it already at 23.976 fps, even if it suffers from combing? That is, aren't the pulldown flags simply (soft) instructions for the 23.976 fps progressive source? Apologies for my ignorance.

[jagabo]

What I don't understand is why once the progressive source, at 23.976 fps, is striped from the pulldown flags that it still needs to be decimated;

Your video is a mix of progressive frames with pulldown flags and interlaced frames. To properly handle it you need to perform pulldown on the progressive sections (creating interlaced frames to match the interlaced sections), then field match back to progressive frames and decimate.

[LouieChuckyMerry]

Your video is a mix of progressive frames with pulldown flags and interlaced frames. To properly handle it you need to perform pulldown on the progressive sections (creating interlaced frames to match the interlaced sections), then field match back to progressive frames and decimate.

Ahhh, thank you, jagabo, now I understand .

I've another question: is there an AviSynth filter to "brighten" the colors? My searching has been fruitless, but I'm probably using the wrong terms. Should I start a new thread?

[KarMa]

http://avisynth.nl/index.php/Tweak

tweak(sat=0) #Any positive number will increase the saturation

[manono]

I've another question: is there an AviSynth filter to "brighten" the colors?

The Sat command within the built-in Tweak filter. Or the Saturate command within the SmoothTweak portion of the third-party SmoothAdjust filter.

tweak(sat=0) #Any positive number will increase the saturation

Any positive number greater than 1 increases saturation. Less than 1 and saturation is decreased. 0 greyscales it.

[jagabo]

Or ColorYUV(cont_u=X, cont_v=Y) where positive values of X and Y increase saturation, negative values decrease it. Tweak(sat=2.0) is the same as ColorYUV(cont_u=256, cont_v=256). Tweak(sat=0.5) is the same as ColorYUV(cont_u=-128, cont_v=-128). ColorYUV allows you to change the two color channels independently when necessary.

[KarMa]

Any positive number greater than 1 increases saturation. Less than 1 and saturation is decreased. 0 greyscales it.

Yeah I was just going by the wiki real quick. Looks like it's wrong, as it gives a float of 0.0 but AvsPmod gives a float of 1.0.

[LouieChuckyMerry]

manono, jagabo, and KarMa: Thank you!

[LouieChuckyMerry]

Oops, seems I've a last question, please. Where would I place the color adjustment filter line in the following script:

### Deinterlace-Match Fields-Decimate ###
LoadPlugin("F:\[0]StandAloneApps\MeGUI-2500(core)2443(data)0.3.5(libs)[Portable]\tools\avisynth_plugin\TIVTC.dll")
AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
TFM(Chroma=False) 
TDecimate()
### Crop ###
Crop(8,0,-8,0)
### Overall Temporal Denoise ###
SMDegrain(TR=1,ThSAD=400,ContraSharp=True,RefineMotion=True,Plane=0,PreFilter=2,Chroma=False,Lsb=True,Lsb_Out=True)
### Resize ###
LinearResize(640,480,Lsb_In=True,Lsb_Out=True)
### Darken-Thin Lines ###
F=DitherPost(Mode=-1)
S=F.FastLineDarkenMod(Strength=24,Prot=6).aWarpSharp2(Blur=4,Type=1,Depth=3,Chroma=2)
D=MT_MakeDiff(S,F).Dither_Convert_8_To_16()
Dither_Add16(Last,D,Dif=True,U=2,V=2)
### Deband ###
GradFun3(Radius=16,SMode=2,StaticNoise=True,Lsb_In=True,Lsb=True)
Dither_Out()

Thanks. Ahhh, and any suggestions for improvement are very welcome .

[LouieChuckyMerry]

Pretty please, where's the best place for the color adjustment line? I know that filter order is important, so I'd like to have it in the correct order. Thanks.

[KarMa]

Pretty please, where's the best place for the color adjustment line?

Don't really think it matters if we are talking about tweak() or ColorYUV(cont_u=X, cont_v=Y) (as jababo suggested). At the end would probably be fine.

[jagabo]

I like to do color/contrast adjustments early in the processing. I'd put it right after TDecimate.

[LouieChuckyMerry]

Thank you very much for the help .

[LouieChuckyMerry]

Back for more . Using jagabo's kind suggestion in Post #3 above:

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
TFM(Chroma=False) 
TDecimate()

I was zipping along with acceptable-if-imperfect results until Season 4, at which point several episodes suffered from more than the occasional short burst of chroma blending (?) and-or minor interlaced chroma artifacts; these exceptions exhibited, for lack of a better term, "shimmering", where all the lines in a frame would seemingly vibrate simultaneously, almost like looking through water. Here's an example.

So I started searching, beginning with vaporeon800's friendly link in Post #4 above, and tried many things that didn't work, including this. Eventually, just because it was also from jagabo and thinking perhaps it referred to the later seasons, I tried this:

Interleave(TFM(Field=1,Mode=1,PP=1),TFM(Field=0,Mode=1,PP=1))
MergeChroma(Blur(0.0,1.0).aWarpSharp(Depth=10))
SRestore(FRate=23.976)

It fixed the shimmering most of the time but also seemed to add other types of artifacts. So I got to thinking (ha ha) and, after much trial-and-error, combined the two scripts thusly:

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Field=1,Mode=1,PP=1),TFM(Field=0,Mode=1,PP=1))
MergeChroma(Blur(0.0,1.0).aWarpSharp(Depth=10))
SRestore(FRate=23.976)

So far in multiple tests this script has not only eliminated the shimmering but also lowered the amount of chroma blending and interlaced chroma artifacts, too. Now, for fear of being labeled greedy, I was wondering if anyone with an intimate understanding of the above script might offer suggestions to improve it. I understand enough so as to not be clueless, but I lack the deeper understanding necessary to achieve artistry. For example, in this test clip there's still bits of chroma blending and interlaced artifacts and I'm wondering if there are any tweaks to the above script to help eliminate them. Or is the chroma blending impossible to fix because it's in the source?

Here's my full working script in case it matters, and thanks for any input:

LoadPlugin("FullPathToDGDecodeNV.dll")
DGSource("FullPathToIndexFile.dgi")
### Deinterlace-Match Fields-Decimate ###
LoadPlugin("FullPathToTIVTC.dll")
AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Field=1,Mode=1,PP=1),TFM(Field=0,Mode=1,PP=1))
MergeChroma(Blur(0.0,1.0).aWarpSharp(Depth=10))
SRestore(FRate=23.976)
### Adjust Color ###
SmoothTweak(Saturation=1.06)
### Crop ###
Crop(8,0,-8,0)
### Gibbs Noise Block ###
Edge=MT_Edge("prewitt",ThY1=20,ThY2=40).RemoveGrain(17)
Mask=MT_Logic(Edge.MT_Expand().MT_Expand().MT_Expand().MT_Expand(),Edge.MT_Inflate().MT_Inpand(),"xor")
MT_Merge(Minblur(),Mask,Luma=True)
### Overall Temporal Denoise ###
SMDegrain(TR=1,ThSAD=200,ContraSharp=True,RefineMotion=True,Plane=0,PreFilter=2,Chroma=False,Lsb=True,Lsb_Out=True)
### Resize ###
LinearResize(640,480,Kernel="Bicubic",A1=-0.5,A2=0.25,Lsb_In=True,Lsb_Out=True)
### Darken-Thin Lines ###
F=DitherPost(Mode=-1)
S=F.FastLineDarkenMod(Strength=24,Prot=6).aWarpSharp2(Blur=4,Type=1,Depth=3,Chroma=2)
D=MT_MakeDiff(S,F).Dither_Convert_8_To_16()
Dither_Add16(Last,D,Dif=True,U=2,V=2)
### Deband ###
GradFun3(Radius=16,ThR=0.55,SMode=2,StaticNoise=True,Lsb_In=True,Lsb=True)
### Preview Source OR Send 16-bit Output To x264 10-bit ###
## Trim()
# SelectRangeEvery(1000,66)
# DitherPost()
Dither_Out()

[jagabo]

We need sample of the original video.

[LouieChuckyMerry]

We need sample of the original video.

D'oh! Too many links, too many links... anyway, thanks for your patience. After seemingly myriad encoding tests it seems that with the working script at the end of Post #18 above there are three main types of artifacts remaining after encoding:

1) S1.E1-FuturamaTestClip[Combing1]: this source clip encoded with the working script leaves a single frame with extreme combing as the door closes on Fry. This type of artifact appears roughly every 1000 frames across all my tests, and is the problem I'd like to solve most.

Edit: after reading (then rereading several more times) the TIVTC/TFM Wiki, then running many a test, it appears that dropping the two "PP=1"s from the Interleave line in the working script, thus returning the post processing mode to its default of 6, has solved problem 1). But I'll run more tests to be certain...

2) S2.E14-FuturamaTestClip[Combing2]: this source clip encoded with the working script leaves a single frame with subtle combing as the man puts his card in the ATM. This type of artifact appears two or three times as frequently as 1) and is much less annoying, but I still wouldn't mind correcting the problem; however, stepping through the source frame-by-frame this seems to be an unfixable flaw in the source and not an issue with (de)interlacing.

3) S3.E14-FuturamaTestClip[ChromaBlending]: this source clip encoded with the working script leaves a frame with chroma (field?) blending as the second baby's arms extend. I'd love to be able to correct this, if possible.

Given my limited knowledge, it seems to me that with skillful tweaks to the working script 1) and 3) are solvable problems. Of course, I'm counting on someone with skills for suggestions, any of which are most gratefully appreciated .

[jagabo]

For the IVTC try the following:

Mpeg2Source("S1.E1-FuturamaTestClip[Combing1].d2v", CPU2="ooooxx", Info=3) 
Interleave(TFM(Field=1,Mode=1,PP=1),TFM(Field=0,Mode=1,PP=1))
TFM(field=1)
TDecimate(cycle=10, cycleR=6)

That removed combing from all three clips but left the aliasing in the second clip and the blended chroma in the third. But those problems are in the source and can't be easily fixed.

The problem in the first clip was two orphaned fields as the door closes. There is no other field with which to match them. The second clip has a frame with aliasing artifacts. The third clip has a frame where the two fields' chroma are blurred together.

[LouieChuckyMerry]

Thank you very much for the suggestion and the explanations, jagabo. Having poked through more of the sources frame-by-frame I think the frames suffering from aliasing and chroma blending are well beyond my current ability (and patience) to correct; however, as single frames at 23.976 fps they also zip by very very quickly, enough so as to be hardly noticeable. I can live with them. It was the sudden frames with drastic combing that I mostly wanted to cure and your update does the trick. I'll run some tests comparing the working script with either

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Field=1,Mode=1,PP=1),TFM(Field=0,Mode=1,PP=1))
TFM(Field=1)
TDecimate(Cycle=10,CycleR=6)

or

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Field=1,Mode=1),TFM(Field=0,Mode=1))
MergeChroma(Blur(0.0,1.0).aWarpSharp(Depth=10))
SRestore(FRate=23.976)

to compare speed and quality, then I'll post the results. Thanks again .

[LouieChuckyMerry]

Happy Friday! After extensive testing it seems that both the above scripts have limitations, so I made a combination that I hope can be perfected with some help.

This script does a really good job with the combing but the chroma blending is a bit lacking:

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Mode=1,PP=1,Field=1),TFM(Mode=1,PP=1,Field=1))
MergeChroma(Blur(0.0,1.0).aWarpSharp(Depth=10))
TFM(Field=1,CThresh=17)
SRestore(23.976)

while the same script minus the "PP=1"s very much improves the chroma blending but fries (Frys? ) the combing

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Mode=1,Field=1),TFM(Mode=1,Field=1))
MergeChroma(Blur(0.0,1.0).aWarpSharp(Depth=10))
TFM(Field=1,CThresh=17)
SRestore(23.976)

Running each script on S1.E1-FuturamaTestClip1 and S3.E14-FuturamaTestClip2 perfectly illustrates the problem.

It seems to me there has to be a way to get the best of both scripts, but for the life of me I can't figure it out. Pretty please, can someone guide me to the promised land . Thank you!

[jagabo]

What about this:

Interleave(TFM(Mode=1,PP=0,Field=1),TFM(Mode=1,PP=0,Field=0)) # reverse field args for bff
TFM(field=0, cthresh=17)
vInverse() # remove small comb artifacts found left in some shots
SRestore(23.976)

The one place that doesn't work is the blended chroma in S3.E14-FuturamaTestClip[ChromaBlending].m1v.

What's different in S3.E14-FuturamaTestClip2.m1v is the way the titles fade in/out. They fade at 60 fields per second, confusing all the TFM calls.

[LouieChuckyMerry]

What's different in S3.E14-FuturamaTestClip2.m1v is the way the titles fade in/out. They fade at 60 fields per second, confusing all the TFM calls.

Ahh, I wondered what was up, thanks for the explanation. A few quick tests with your updated script are very promising; I'll run extended tests and post results. Many thanks .

Happy Saturday

[LouieChuckyMerry]

After running a couple SelectRangeEvery(66,1000)'s for each season, your last update appears to function as well as possible given the source; the combing and chroma blending are minimized, a fine combination of the two earlier scripts. Thank you! I'll run a full episode overnight then reassess. The only issue remaining is the opening title sequences. As you were kind enough to point out, their fade ins-outs are causing trouble. Adding "CThresh=17" fixed this for most of the episodes, including S3.E14-FuturamaTestClip[ChromaBlending] (which happens to include titles). However, S4.E13-FuturamaTestClip[Titles2] needed "CThresh=23" in order to eliminate the combing bursts in the crowd scene, and for S3.E2-FuturamaTestClip[Titles1] I was unable to remove the title sequence combing when the trucker shakes his hips left-to-right no matter how high I set the "CThresh". This leaves me with two questions:

1) Why does raising the "CThresh" work for one clip but not the other? Is it just luck that some of the TFM calls for the 60 fields per second sections leave "fixable" frames?

2) Is there any way in the script to differentiate between the 60 fields per second sections and the rest? That is, can an if-then type statement be added to the script, something along the lines of "If 60 fields per second, then TFM this way. If otherwise, then TFM this other way?"

Now to crawl into bed. Thank you thank you thank you...

[jagabo]

Note that the titles1 test clip contained a file called titles2. It has "hip shaking" so I assumed it was the titles1 clip you referred to. I'm not seeing any residual combing in that shot with the earlier script. But it's BFF where most of the others have been TFF. Swapping the field=N arguments might help. Or you can change the Inteleave() line to:

GetParity() ? Interleave(TFM(Mode=1,PP=0,Field=1),TFM(Mode=1,PP=0,Field=0)) : Interleave(TFM(Mode=1,PP=0,Field=0),TFM(Mode=1,PP=0,Field=1))

The Interleave(TFM...TFM) is essentially a smart bob doubler for telecined film, where the field order is specified by the two field=N statements, not AviSynth's field order flag (parity). The above line will work with both TFF and BFF sources.

1) Why does raising the "CThresh" work for one clip but not the other?

Cthresh determines how much combing TFM has to see before it decides a frame is interlaced. The higher you set it the more combing it has to see before it deinterlaces.

Is it just luck that some of the TFM calls for the 60 fields per second sections leave "fixable" frames?

Basically, yes.

2) Is there any way in the script to differentiate between the 60 fields per second sections and the rest?

All of the clips have been 60 fields per second. The issue is whether consecutive fields really contain changes at every field.

That is, can an if-then type statement be added to the script, something along the lines of "If 60 fields per second, then TFM this way. If otherwise, then TFM this other way?"

I don't see any easy way to detect this.

[LouieChuckyMerry]

Note that the titles1 test clip contained a file called titles2. It has "hip shaking" so I assumed it was the titles1 clip you referred to. I'm not seeing any residual combing in that shot with the earlier script.

I fixed the name, thanks. With the script

AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Mode=1,PP=0,Field=1),TFM(Mode=1,PP=0,Field=0))
TFM(Field=0,CThresh=17)
Vinverse()
SRestore(23.976)

I get 3 contiguous combed frames the second time his hips move left, just as "Supervising Director Rich Moore" fades in, then four contiguous combed frames the third time his hips move left, just as "Supervising Director Rich Moore" fades out. It's the same if I swap the Field=N arguments or change the Interleave line to your suggestion. Was the earlier script that worked for you the one above? Here's my result, in case what I'm calling residual combing isn't .

Cthresh determines how much combing TFM has to see before it decides a frame is interlaced. The higher you set it the more combing it has to see before it deinterlaces.

I read the TIVTC/TFM Wiki repeatedly, and the CThresh section rerepeatedly, and it still confuses me. It seems to me that setting "CThresh=1" would solve any problem because then basically all combing would be eliminated, but that's not the case. I'll read it some more times...

2) Is there any way in the script to differentiate between the 60 fields per second sections and the rest?

All of the clips have been 60 fields per second. The issue is whether consecutive fields really contain changes at every field.

Is there a way to determine whether consecutive fields really contain changes at every field, or is that question absurdly naive?

[jagabo]

I think you're misinterpreting aliasing with combing. When the Rich Moore titles fade in and out the original video has been bob deinterlaced leaving lots of aliasing artifacts. The script does nothing to try to reduce those. You can try using QTGMC(InputType=2) after SRestore() to reduce the aliasing. But it will cause damage to other parts of the video. I think it's going to be hard detecting frames with those aliasing artifacts and fixing only those.

[LouieChuckyMerry]

Yep, I definitely need to get the artifacts straight . In my dozen SelectRangeEvery(66,1000) tests I only noticed the aliasing twice, and for just a quick couple-four frames at that. So, I'll run some Trim tests with the opening credits and see how bad it is; if it's bearable, great, if not I'll probably pester you some more. As of now, the script

LoadPlugin("SourcePath.dll")
DGSource("SourcePath.dgi")
### Deinterlace-Match Fields-Decimate ###
LoadPlugin("F:\[0]StandAloneApps\MeGUI-2500(core)2443(data)0.3.5(libs)[Portable]\tools\avisynth_plugin\TIVTC.dll")
AssumeTFF()
TFM(Chroma=False,PP=0) 
AssumeBFF()
Interleave(TFM(Mode=1,PP=0,Field=1),TFM(Mode=1,PP=0,Field=0))
TFM(Field=0,CThresh=23)
Vinverse()
SRestore(23.976)
### Adjust Color ###
SmoothTweak(Saturation=1.06)
### Crop ###
Crop(8,0,-8,0)
### Gibbs Noise Block ###
Edge=MT_Edge("prewitt",ThY1=20,ThY2=40).RemoveGrain(17)
Mask=MT_Logic(Edge.MT_Expand().MT_Expand().MT_Expand().MT_Expand(),Edge.MT_Inflate().MT_Inpand(),"xor")
MT_Merge(Minblur(),Mask,Luma=True)
### Overall Temporal Denoise ###
SMDegrain(TR=1,ThSAD=200,ContraSharp=True,RefineMotion=True,Plane=0,PreFilter=2,Chroma=False,Lsb=True,Lsb_Out=True)
### Resize ###
LinearResize(640,480,Kernel="Bicubic",A1=-0.5,A2=0.25,Lsb_In=True,Lsb_Out=True)
### Darken-Thin Lines ###
F=DitherPost(Mode=-1)
S=F.FastLineDarkenMod(Strength=24,Prot=6).aWarpSharp2(Blur=4,Type=1,Depth=2,Chroma=2)
D=MT_MakeDiff(S,F).Dither_Convert_8_To_16()
Dither_Add16(Last,D,Dif=True,U=2,V=2)
### Deband ###
GradFun3(Radius=16,ThR=0.55,SMode=2,StaticNoise=True,Lsb_In=True,Lsb=True)
### Preview Source OR Send 16-bit Output To x264 10-bit ###
## Trim()
# SelectRangeEvery(1000,66)
# DitherPost()
Dither_Out()

does a very nice job with everything but the titles. The only combing artifacts remaining are in the source and the chroma blending is reduced to a few odd frames here and there that are hardly noticeable at normal playback speed. As always, thank you very much for the help, I really appreciate it .

Edit: I remembered that I had a question. Before I learned that it was aliasing, I was using CThresh to eliminate what I thought was residual combing. And it worked! Why would raising the CThresh to 23 eliminate some of the aliasing that I thought was combing?