How to get rid of "little blocks" effect of DVD-Video

[provato]

Hi there, and first of all excuse my ignorance of a proper name for the effect in the video, that I want to remove. Here is a picture that shows the little blocks around edges of a cartoon video (this video screenshot is from an mp4 DVDRip of mine):

Please note that I use avisynth for different filters/corrections, so I would appreciate if someone knows an avisynth line for correcting this "blocking" effect. (I have already UNsuccessfully used x264's deblocking...)

[hech54]

(this video screenshot is from an mp4 DVDRip of mine)

You created this MP4 file from a DVD?

[provato]

(this video screenshot is from an mp4 DVDRip of mine)

You created this MP4 file from a DVD?

Yeah, why? Is it so bad? This is just a screenshot from the mp4 DVDRip

[Ai Haibara]

Well, ripping the DVD alone won't get you an .mp4 video. For that, you have to rip and convert - and the thing about that is you're most likely to lose quality while converting.

That said, I think I remember Disney's Darkwing Duck DVDs didn't look as good as I thought they would... maybe VHS quality in some places. I'll have to remind myself to watch them again, sometime.

[provato]

Well, ripping the DVD alone won't get you an .mp4 video. For that, you have to rip and convert - and the thing about that is you're most likely to lose quality while converting.

That said, I think I remember Disney's Darkwing Duck DVDs didn't look as good as I thought they would... maybe VHS quality in some places. I'll have to remind myself to watch them again, sometime.

Thanks for the info, but let's stick to the effect I would like to eliminate.
I hope that someone has encountered this before and got rid of it adequately. If so, please let me know

Also, let me inform you that this effect is also present in DVD playback, so it's not my conversion that did this.

[LMotlow]

Might be better to learn what "rip" means, along with the term "re-encode". You apparently ripped + re-encoded to a lower bitrate. What was wrong with the original encoded DVD that it required re-encoding into a new container at lower bitrates?

[provato]

Might be better to learn what "rip" means, along with the term "re-encode". You apparently ripped + re-encoded to a lower bitrate. What was wrong with the original encoded DVD that it required re-encoding into a new container at lower bitrates?

Oh my....
Guys please, if I want to re-encode or whatever, it's my right to do for personal purposes, even though I know it's lower quality.
I want it to be lower quality, but without this "blocking" effect.
Is it so hard for you to answer me this question, without a lecture about what "I should have done" to the DVD????

so:

YES, I want to re-encode to mp4
YES, this effect is also present in the original-untouched video
YES, I will be satisfied with the lower quality, if this effect is eliminated
NO, I don't want to consider something else than mp4 converting

And I'm sorry I used the word "DVDRip". My mistake, won't happen again.

[LMotlow]

If block noise is in the original, you need an anti-block filter. Macroblocks are very difficult to remove from poorly encoded video. The posted image also shows mosquito noise and other artifacts on edges. Without a short sample from the original (which means an unprocessed short cut from the rip, not from the rip+conversion), it would be tough to say what filters to use.

Deblocking: DeBloock_QED or the original DeBlock. http://avisynth.nl/index.php/External_filters#Deblocking
mosquito noise: anyone's guess. Try TComb. This looks like a tape captured to DV (yep, that's a guess). There are dot crawl filters around that can help mtoo, nbut they're rather destructive. http://avisynth.nl/index.php/External_filters#Deringing_.26_Mosquito_Noise
Other edge artifacts: HQdering, DeHalo_Alpha. http://avisynth.nl/index.php/External_filters#Dehaloing
A longer listing, many filters by type: http://avisynth.nl/index.php/External_filters

Why are you deinterlacing DVD? This appears to be movie-based animation, which is usually telecined.

You can't have low-bitrate low-quality and eliminate all the artifacts entirely at the same time. Low-quality encoding causes it in the first place. You'll also have banding effects unless you apply some anti-banding like GradFun2dbMod and reapply fine grain to fill in the "details" removed by over-filtering or lost thru crummy re-encoding.

[manono]

I don't see any blocking. Dot crawl, maybe, and mosquito noise. As LMotlow mentioned, a sample from the DVD showing the problem will help.

[provato]

Thank you, thank you, thank you for this great answer!!!

If you have some time, I will upload later today a DVD-Video sample (untouched, telecined, I'll only use trim on the original DVD-Video)
Please try some filters on the sample and I would appreciate if I can learn a few things from your chosen process (in simple words please because I am a beginner)

I use detelecine (TIVTC) because I simply like it better deinterlaced, no particular reason.

[provato]

Here is the DVD-video sample

(sorry for double post, please merge if needed)

[LMotlow]

I don't see any blocking. Dot crawl, maybe, and mosquito noise. As LMotlow mentioned, a sample from the DVD showing the problem will help.

Block noise and hard edges in the purple background, mostly upper left. Probably easier to see in a video sample.

Oops, I just noticed the sample has been posted.

[LMotlow]

I use detelecine (TIVTC) because I simply like it better deinterlaced, no particular reason.

Just had a brief look at the sample, haven't had time to get more detailed yet.

But here's a tip: the sample is progressive video with pulldown (telecine) applied. If you used TIVTC the result is progressive 23.976, not interlaced. If you follow that with QTGMC, the result plays at 49.95 fps and every frame has a duplicate.

dot crawl filters are usually applied before IVTC. Sometimes you need a little more afterwards.

EDIT: DGindex on the VOB reports a field order change somewhere in the video. Will have to check that out later.

[jagabo]

Maybe you could start with something like this:

Mpeg2Source("DD_Sample.d2v", CPU2="xxxxxx", Info=3) 
CheckMate(thr=10, max=20, tthr2=0) # reduce dot crawl
TFM(pp=0) 
TDecimate() 
Vinverse()
ChubbyRain2(th=4) # reduce rainbows
MergeChroma(aWarpSharp(depth=5), awarpsharp(depth=10)) # sharpen luma and chroma
McTemporalDenoise(settings="low") # reduce noise

Mpeg2Source's deblocking is usually removes too much detail but since this is cartoons it's not so bad. You could replace it with DeblockQED() for less deblocking with more detail. Keep the DCT deringing though.

There's one problem I didn't address: sharp vertical edges have ghosts a few pixels to the left and right.

[provato]

Maybe you could start with something like this:

Mpeg2Source("DD_Sample.d2v", CPU2="xxxxxx", Info=3) 
CheckMate(thr=10, max=20, tthr2=0) # reduce dot crawl
TFM(pp=0) 
TDecimate() 
Vinverse()
ChubbyRain2(th=4) # reduce rainbows
MergeChroma(aWarpSharp(depth=5), awarpsharp(depth=10)) # sharpen luma and chroma
McTemporalDenoise(settings="low") # reduce noise

Mpeg2Source's deblocking is usually removes too much detail but since this is cartoons it's not so bad. You could replace it with DeblockQED() for less deblocking with more detail. Keep the DCT deringing though.

There's one problem I didn't address: sharp vertical edges have ghosts a few pixels to the left and right.

Wow! Really really good elimination of those "blocky edges" but I'm afraid these filters produce some strange artifacts on some frames:
(Note that I cannot use vinverse.dll since I use 2.5.x avisynth)

EDIT: Chubbyrain2 produces these artifacts

[jagabo]

I didn't get that:



Maybe a version difference.

[LMotlow]

I didn't get that

I didn't either. Didn't use chubbyrain2, I used dfttest and cnr2. But the mottling might be from wrong version of masktools ? ?



There's still some schmutzy stuff in backgrounds. Probably have to go to dithering to smooth it out. Could use maybe some gamma or midtone work, too. Looks a little dim. Jagabo's sample is snappier. There's also some gray edge ghosting.

I keep seeing png images looking darker than jpg in this forum. The bottom image (attached) is jpg. Origianl post above is png.
Whatever.....

[provato]

I didn't get that

I didn't either. Didn't use chubbyrain2, I used dfttest and cnr2. But the mottling might be from wrong version of masktools ? ?

Could you please provide a dfttest + cnr2 avisynth example, to test on my script?
I use MaskTools v2 - alpha 2

[jagabo]

I use MaskTools v2 - alpha 2

I'm pretty sure that's what I'm using too. Attached is the version of ChubbyRain2 I'm using (AVSI in ZIP).

[LMotlow]

Working with jagabo's idea now, but just got interrupted. Back soon.

[provato]

I'm pretty sure that's what I'm using too. Attached is the version of ChubbyRain2 I'm using (AVSI in ZIP).

yes, we use the same version of chubbyrain2 too. I also checked the checksums (CRC32) to be sure.

EDIT: Please correct me if I'm wrong: The main problem seems to be this "dot crawl" that appears as small blocks/rectangles/squares around edges, mostly on yellow. I figured this out because using checkmate or decrawl or TComb eliminates most of the problem. After that is done, we need a filter to de-noise the edges without too much blurring.
Correct?

[LMotlow]

Don't know why there's discoloration with chubbyrain2. I've used it often with no problems. You might try this code, which has often been used for color corruption and rainbows:

Cnr2("xxx",4,5,255)
FFT3DFilter(bt=3,plane=3)

EDIT: Please correct me if I'm wrong: The main problem seems to be this "dot crawl" that appears as small blocks/rectangles/squares around edges, mostly on yellow. I figured this out because using checkmate or decrawl or TComb eliminates most of the problem. After that is done, we need a filter to de-noise the edges without too much blurring.
Correct?

Well...mmm, not quite. There's more than one major problem. I Count, block noise, banding (hard edges instead of smooth gradients), dot crawl, halos, edge ghosts being some of 'em. Looks as if whoever made the dvd did too much filtering, then too much sharpening. And I don't know what else. Film-based source and tape are grainy of course, but a lot of the "detail" is in that grain. Looks like someone tried to remove all the grain. Bad move. Tehn a bitrate was used that was too0 low for this work, or maybe something like a quanty parameter or whatnot wasn't set bright, so you have all that mosquito noise (ccompression artifacts. Or maybe just a crummy encoder.

The dot crawl isn't mostly in yellow, it's mostly everywhere. The image below is from the original VOB. Shows a lot of the problems mentioned. Also note, some of the vertical lines are just dot crawl, period, no solid line. The image at the bottom is after some of the filtering I've done so far, but I keep getting interrupted today. I'll keep at it.

[jagabo]

Please correct me if I'm wrong: The main problem seems to be this "dot crawl" that appears as small blocks/rectangles/squares around edges, mostly on yellow. I figured this out because using checkmate or decrawl or TComb eliminates most of the problem.

Yes. You will see them at the edges of highly saturated colors. It's the result of the chroma subcarrier not being completely separated from the luma.

With VHS you can usually use something like BicubicResize(width/2,height).BicubicResize(width,height) to reduce dot crawl because the underlying video isn't sharp. But this video is too sharp for that (it will get get noticeably blurrier).

After that is done, we need a filter to de-noise the edges without too much blurring.
Correct?

Yes, after reducing the dot crawl artifacts there are still some rainbow artifacts and misc. noise. Hence the ChubbyRain2 and MCTD.

[LMotlow]

With VHS you can usually use something like BicubicResize(width/2,height).BicubicResize(width,height) to reduce dot crawl because the underlying video isn't sharp. But this video is too sharp for that (it will get get noticeably blurrier).

Not so much "sharp" as over-sharpened, is what I'm seeing. Not much fine textural detail any more, just clumpy remnants and lots of fractured and raggedy lines, some line twitter, ugly scene breaks. Mosquito noise + ghosty edges sure looks like tape-to-DV to me. But it's anybody's guess.

Made 3 or 4 tries, saw different problems in different segments. 23.976 progressive mp4, AAC audio.

[provato]

...I Count, block noise, banding (hard edges instead of smooth gradients), dot crawl, halos, edge ghosts being some of 'em....

...It's the result of the chroma subcarrier not being completely separated from the luma...

...Not so much "sharp" as over-sharpened, is what I'm seeing. Not much fine textural detail any more, just clumpy remnants and lots of fractured and raggedy lines, some line twitter, ugly scene breaks. Mosquito noise + ghosty edges sure looks like tape-to-DV to me. But it's anybody's guess...

Thank you LMotlow and jagabo, I guess I have to study harder on these notes of yours I quoted, because I can see the same tape-to-DV damage on many (disney mostly) DVDs. I think I can recognize dot crawl now (and I thank you both again for that), but I need to learn how to recognize different types of noise too.

One more thing I wanted to say (please don't burn me for this!) is that I'm not a perfectionist in the level that you are. I mean that I don't mind a few rainbow artifacts or slight discolouration every few frames on the video, because it's a fast-paced animation and these minor mistakes are not highly noticeable in playback. Of course I have to remove edge ghosting, dot crawl and most of the edge noise, so that's where I'll be focusing.

PS: Awesome job on that sample LTMotlow, looks like you have corrected it completely frame-by-frame!!!

[LMotlow]

PS: Awesome job on that sample LTMotlow, looks like you have corrected it completely frame-by-frame!!!

Well....thanks, but not "completely" fixed. It came down to making two versions with two slightly different sets of filters, then picking out pieces from both and encoding together. Kind of like sorting out the new eggs in the morning (yep, I know everybody here has done that, right?). A few tricks I learned here from jagabo and a few obsessive types who don't seem to be around any more. Some tricks you really don't like to use because of side effects, but...do what you can, I guess. Jagabo's earlier idea looked OK to me, mainly one had to figure a way to handle the edge problems. I ran a couple of standalone versions of filters used internally by MCTemporalDenoise.

Sorry to ramble -- the first A.M. coffee is still soaking in. Will try to get scripts and sample pics later. Maybe the critics can tweak what I did. I have questions myself.

[jagabo]

I noticed there's an orphaned field 25.15 seconds into the clip. Not using the post processor in TFM(pp=0) lets that through as an interlaced frame. For what it's worth, here's the result of my script. I removed pp=0 from TFM and added dehalo_alpha(rx=4, ry=2) after vinverse. The latter softens the picture quite a bit so some more sharpening may be in order.

[provato]

Wow really good work, I could never imagine this level of correction could ever be reached.
Moreover, as a matter of fact, I prefer sharper edges and overall too (like jagabo mentioned) even though it leaves a small amount of edge ghosting.
So my questions are: 1. Can you show me a part that has "halos" so that I can recognize it in an animated film? and 2. Is Vinverse worth enough, so that I should consider upgrading to avs 2.6.0? (since I hardly ever come across residual combing after IVTC)

[jagabo]

1. Can you show me a part that has "halos" so that I can recognize it in an animated film?

Look at LMotlow's first image in post #22. Jusb below the croc's right elbow there is a vertical black line between two different shades of green. But rather than simply transitioning from dark green to black line to light green, you can see a brighter shades of green just to the left and right of the black line (8x point resize):



Even worse, the bright halos have dark halos to their sides. You can see these oversharpening artifacts throughout the video. They are caused by the routine sharpening filter used in analog tape playback. And probably exacerbated by digital sharpening during or after capture.

2. Is Vinverse worth enough, so that I should consider upgrading to avs 2.6.0? (since I hardly ever come across residual combing after IVTC)

Vinverse here isn't too critical. In a few frames it blurred away some residual ghosting in the chroma channels:

before vinverse:


after vinverse:


If you zoom into the images you'll see alternating green and purple horizontal lines in and above the characters right hand. vinverse blurred the lines together. There are a few frames where the inverse telecine resulted in one field being slightly lighter than the other. Vinverse blended those together too. If you leave vinverse out noise filtering at later steps will probably blur in a similar way anyway.

[LMotlow]

I noticed there's an orphaned field 25.15 seconds into the clip. Not using the post processor in TFM(pp=0) lets that through as an interlaced frame.

Now that's weird. I ran your script verbatim yesterday but didn't get an interlaced frame. What I did get at 25.15 with your entire script running is this, with the rough lines:


I got the same result removing pp=0 and removing vInverse. In my case I ran two scripts. The first of my scripts used a different MPEG2Source statement. I got this for the same frame (with my entire script running):


I also ran a second script, same MPEGSource but stronger cleaners:


The frame in the original VOB is interlaced (telecine). This is your opening code:

Mpeg2Source("DD_Sample.d2v", CPU2="xxxxxx", Info=3)
CheckMate(thr=10, max=20, tthr2=0) # reduce dot crawl
TFM() 
TDecimate() 
Vinverse()

This is the opening code I used in both of my scripts:

aud=WAVSource("F:\provato\DD_Sample.wav")
vid=MPEG2Source("F:\provato\DD_Sample.d2v")
AudioDub(vid,aud)
TComb()
TFM(order=1).TDecimate()

I don't think the audio lines or order=1 affects this business. If I just use "MPEG2Source("F:\provato\DD_Sample.d2v")" without importing the audio I get the same results. But the way I created the d2v might have made a difference. When I opened the VOB in DGIndex I got a warning that a field order transition was detected. DGindex offered to correct it, so I said yes. I never did find the spot where that transition occurred, but I gave up looking kinda early.

I can try running DGIndex without the DGIndex correction. Maybe that would give me different results. Thanks for that tip about pp=0, I'll read up on it again.