Improving a PAL video movie

[carlmart]

I have a swedish PAL DVD that I would like to try improving the image, which is rather poor. This is what I get with MediaInfo from the demuxed file:

Complete name : C:\Temp\VideoFile.m2v
Format : MPEG Video
Format version : Version 2
File size : 3.98 GiB
Duration : 50mn 49s
Overall bit rate : 11.2 Mbps

Video
Format : MPEG Video
Format version : Version 2
Format profile : Main@Main
Format settings, Matrix : Default
Duration : 50mn 49s
Bit rate mode : Variable
Bit rate : 10.8 Mbps
Nominal bit rate : 9 800 Kbps
Width : 720 pixels
Height : 576 pixels
Display aspect ratio : 4/3
Frame rate : 25.000 fps
Standard : PAL
Colorimetry : 4:2:0
Scan type : Interlaced
Scan order : Top Field First
Bits/(Pixel*Frame) : 1.039
Stream size : 3.82 GiB (96%)


I want to keep it PAL, but I would like to enhance it somehow, improve the resolution if possible. Or at least cheat the viewer that it was improved, as most upscalers do. Is there anything you would advice me to try

[DB83]

Resolution-wise = nowhere to go. DVD has a maximum setting and this disk appears to hit that.

[davexnet]

Also the original poster doesn't say what he wants to do with his re-encode - Create another DVD? Something else?

Also, perhaps post a small section so we can what the problem is.

[manono]

Yep, carlmart, you should know by now that samples are always needed.

Complete name : C:\Temp\VideoFile.m2v
Format : MPEG Video
Format version : Version 2
File size : 3.98 GiB
Duration : 50mn 49s
Overall bit rate : 11.2 Mbps

Did you already reencode this thing? You weren't planning on authoring that, were you?

[carlmart]

Sorry, I was awaiting for instructions on what to do.

If the results are good I am willing to re-encode it, of course, re-author it for my personal use.

I would be willing to upload to Mediafire a small section, of course. I just don't remember how that was done, only that DGindex was used.

How large a part?

[manono]

Open a VOB in DGIndex and use the [ and ] buttons to isolate a small section. Then File->Save Project and Demux Video. Upload the resulting M2V. How large a part? Something that illustrates the problem(s). Something with steady movement. 10-20 seconds should be enough.

[carlmart]

OK. Here it is:

http://www.mediafire.com/?zbfxrchijzikvzd

Please let me know if you need a different part with specific characteristics.

[manono]

No, that's a good sample. I suppose it's some sort of a TV show and they stuck too many episodes on the DVD because the bitrate is too low (not much over 3,000) and the average Q is way too high, even though a good quantisation matrix was used. It has blocking which I'd treat first (DeBlock_QED or some such), followed by something to clean up all that noise (FFT3DFilter or your favorite denoiser/smoother), possibly followed by a sharpener, but maybe not (LimitedSharpenFaster or whatever). If you think the colors are too 'dull', maybe juice up the saturation (Tweak). Others might have other suggestions, but that's what I use with similar sources.

[davexnet]

Is that Prime Suspect? Every time I've caught it on PBS it looks like crap.
Perhaps it was filmed in 16mm, or something like that.

[carlmart]

OK. It looks promising, particularly because there are two filters I never tried before. So I might need some help getting the right results from them.

What is the blocking and how to I work on it, which are the variables?

I don't have a favorite denoiser/smoother, so I am open to suggestions. What would be a good starting point for say FFT3DFilter?

Yes, it is Prime Suspect. The first seasons look quite bad indeed, and I don't think they were shot in 16mil. But you might be right.

[DB83]

Is that Prime Suspect? Every time I've caught it on PBS it looks like crap.
Perhaps it was filmed in 16mm, or something like that.

Yes, Prime Suspect. Starring her majesty, Dame Helen Mirren.

Now from where I sit, the picture looks ok - it is a few years old after all. Maybe also there is translation (PAL >> NTSC) issue when it goes over the pond.

The blockiness. Could that be caused by too low a bit-rate which is not representative of the original data. ?

But we also do not know the OP's definition of 'rather poor'. One other thought how close is PAL-I to PAL-M which I thought was similar to NTSC ?

[carlmart]

Please disregard completely any PAL-M thoughts or ideas you might have. This seems to be a common place every time I bring a video problem here, and it never has anything to with it.

To start with this DVD is European PAL, released in Sweden. It even has subtitles in Swedish that I will happily eliminate in this new authoring.

Second, PAL-M was never, repeat never, part of the home video marketing, not even in VHS times. All releases were ALWAYS in NTSC, not modified NTSC, but pure NTSC. Fortunately. PAL-M was only a broadcasting distortion which now, in satellite times, is way past by.

The original poster defines as rather poor an image with grain, or noise, or muted colors, or anything that might affect the illusion of tridimensionality that every cinematographer will look for on his lighting. I am a film & video professional, coming from film times, so I do know what film and lenses are capable of, particularly one from the mid-90s, which is when this specific film was released. If it was shot in 16mm this would be valid for the average screen size we use today, which is 42".

Whatever I can do to get back some film quality that was lost in the film transfer, which is probably the case, or "cheat as if it did", would make me happy. Am I asking too much?

[DB83]

Ok. Thanks for the fill in.

Dare I ask why you got that from Sweden when the whole lot is available in PAL from the UK ?

The other thing which I do not understand if your sample is from the video in the original media info grab then why is the data rate so significantly different ?

I would have thought that even in mid 90's interior studio was video based with only some location shooting on film. And I would have thought that 16mm would show a whole lot more grain than this example shows.

[carlmart]

A Brazilian friend of mine lives in Stockholm, and once I told him about that series and how I considered it one of the best there is. And one day he found a DVD collection and brought it to me.

But I was quite frustrated with what I expected to be a good quality image, and it was not. Now maybe I can do something about it. I was expecting it to be released in HD, with new mastering, but I don't think that will happen.

In the '90s, all of it, very few shows were being shot in video because they were expecting HD soon, and film was more "ready" for that, also with the right aspect ratio.

I have very few doubts that the image problems we see now are due to a poor telecine, and I would like to see if something can be done to improve on that. It's only the first seasons, up to 4, that hold that poor quality. Things get better later. Unfortunately only the last two seasons were 16:9.

Just if don't know it, but Law & Order was shot in super-16mm for a few seasons, though I don't know when that stopped.

With modern film lenses and lighting, grain is really a thing of the past in 16mm.

[manono]

Is that Prime Suspect? Every time I've caught it on PBS it looks like crap.

Every BBC show I've ever seen on PBS here in the US or on NTSC DVD has been field-blended from a poor standards conversion and blurry during movement. His is PAL, so it doesn't have that particular problem.

What is the blocking and how to I work on it, which are the variables?

Go over it frame by frame. In the parts where there's the greatest movement the screen breaks up into little blocks, little squares. As for DeBlock_QED, the default setting should be fine. The stronger you make it the more detail you lose. You use a deblocker before any cropping or resizing or other filtering.

For the denoising, the doc's best settings suggestion for FFT3DFilter is good (but slow). Play with the sigma value until it looks as you want it. The higher the value the more detail you lose, but the more of that noise will disappear. 1.5-2.5 should be good. You may or may not want to sharpen within FFT3DFilter. I don't think its sharpener is all that good and prefer to do it separately.

[carlmart]

[QUOTE=manono;2085740]

Go over it frame by frame. In the parts where there's the greatest movement the screen breaks up into little blocks, little squares. As for DeBlock_QED, the default setting should be fine. The stronger you make it the more detail you lose. You use a deblocker before any cropping or resizing or other filtering.

Default is <DeBlock_QED()>?

For the denoising, the doc's best settings suggestion for FFT3DFilter is good (but slow). Play with the sigma value until it looks as you want it. The higher the value the more detail you lose, but the more of that noise will disappear. 1.5-2.5 should be good. You may or may not want to sharpen within FFT3DFilter. I don't think its sharpener is all that good and prefer to do it separately.

Removegrain loads fine, but it doesn't seem to do anything.

I couldn't make FFT3DFilter to work, neither Convolution3D. I'm getting script errors.

Right now my script loaded on AvsP is:

import("C:\video\AviSynth 2.58\plugins\deblock_QED.avs")
#loadplugin("C:\video\AviSynth 2.58\fft3dfilter.dll")
#loadplugin("C:\video\AviSynth 2.58\Convolution3D.dll")

MPEG2Source("c:\temp\VideoFile.d2v") 
DeBlock_QED(quant1=30,quant2=51)
#FFT3DFilter(sigma=2, sharpen=0.3, interlaced=true)
#Removegrain (mode=17)
#Convolution3d (0, 32, 128, 16, 64, 10, 0)


LSFmod(smode=5)

What I'm doing wrong?

[manono]

Did you read the FFT3DFilter document:

Features and limitations

.
.
.
You MUST put FFTW3.DLL file from this package (not fft3dfilter.dll) to some directory in path (for example, C:\WINNT\SYSTEM32).
Filter will NOT work without it!

Default is <DeBlock_QED()>?

Yes, I mostly just use Deblock_QED() unless I have particularly bad blocking, in which case I start strengthening it.

[carlmart]

Did you read the FFT3DFilter document:

Features and limitations

.
.
.
You MUST put FFTW3.DLL file from this package (not fft3dfilter.dll) to some directory in path (for example, C:\WINNT\SYSTEM32).
Filter will NOT work without it!

I knew you would ask that!

Yes, I did follow those instructions, also updated Masktools to 2.0a48.

But why I can't load Convolution3D either?

Yes, I mostly just use Deblock_QED() unless I have particularly bad blocking, in which case I start strengthening it.

As you can see, I also try those variables on my script, comparing to (). And I don't seem to see any difference.

[jagabo]

Did you read the FFT3DFilter document:

Features and limitations

.
.
.
You MUST put FFTW3.DLL file from this package (not fft3dfilter.dll) to some directory in path (for example, C:\WINNT\SYSTEM32).
Filter will NOT work without it!

I knew you would ask that!

Yes, I did follow those instructions, also updated Masktools to 2.0a48.

Are you running a 64 bit version of Windows? Then the dll goes in C:\Windows\SysWow64.

By the way, try using Mpeg2Source() with its deblocking and deringing feature:

Mpeg2Source("filename.d2v", CPU=6) #deblock and dering Y, U, and V

[carlmart]

Are you running a 64 bit version of Windows? Then the dll goes in C:\Windows\SysWow64.

No, plain XP SP3 with 32bit.

By the way, try using Mpeg2Source() with its deblocking and deringing feature:

Mpeg2Source("filename.d2v", CPU=6) #deblock and dering Y, U, and V

Now this did bring an improvement! What did we do?

[jagabo]

From the DgMpegDec manual:

cpu: 0 to 6 (default: 0)
Post-Processing Quickset Options.
(Y=luma, C=chroma, H=horizontal, V=vertical)
- 0: DISABLE POST-PROCESSING
- 1: DEBLOCK_Y_H
- 2: DEBLOCK_Y_H, DEBLOCK_Y_V
- 3: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H
- 4: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V
- 5: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V, DERING_Y
- 6: DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V, DERING_Y, DERING_C

[carlmart]

OK. Things improved a lot just using this:

MPEG2Source("c:\temp\VideoFile.d2v", CPU=6) 
LSFmod(smode=5)

What else should I try?

[jagabo]

That video already has over sharpening halos. Try one of the dehalo filters like BlindDehalo3(). Or even a Blur(0, 1.0).

[carlmart]

Should I lose LSFmod then?

What I remember from BlindDehalo is that it took ages to convert.

Blur didn't make much of a difference.

Another thing: does someone know if AvsP has some Java running when it's on?

I'm getting some weird things on the upper half of the screen, where the script is, as it goes transparent to whatever is under it.

[jagabo]

Mpeg2Source("VideoFile_test.demuxed.d2v", CPU=6)
BlindDehalo3(last, rx=3, ry=1, strength=200)

before:


after:


Note the transitions from the white wall to the black suits and back to the white wall. In the before image you can see strong halos. In the after image they are mostly gone. Yeah, it's pretty slow.

[carlmart]

The image also lost a lot of contrast, which I'm not sure I want. The image goes too soft, which is one thing I want to go against.

Now it's late night, so I won't be able to make further tests.

Are there other dehalo filters that may be tried? What to do with the contrast?

[jagabo]

No, the image didn't lose contrast. The forum software is doing that when it makes a thumbnail. Click on the two image to open them full size in their window. You can try using LSFMod() afterward to restore some of the sharpness.

[carlmart]

Yes, you're right. Contrast is preserved.

But the result is too soft. Certainly a result of what the filter does to the border. And it does look as if I had put a fog filter in front of it.

The problem is my quest is just in the opposite direction. What I look for is more "apparent" resolution, not less. I am looking for ways to cheat on that, if that's possible. Which possibly isn't considering the original.

I was even thinking of adding a bit of grain to see what happened.

[jagabo]

You have to remove the halos before you can sharpen the video. Otherwise you will simply enhance the halos. You can play around with the dehalo parameters if you think they're too strong.

[carlmart]

Of course about removing the halos before sharpening. That at least I have learned from you, guys!

You are right about the dehalo correction: it does improve on things and does not get as soft as it looked on the image.

Getting better and better!

I am using LSFmod(smode=5) after BDH3, but I wonder if there's something else I could try to get some resolution back.

I mentioned adding grain, but I am not sure if that is what should be done.