Archiving masters in mjpeg vhs source

[mammo1789]

Hi guys is it "smart" to keep copy of captured video in mjpeg format. I made some comparation between lagarith, dv and pic video mjpeg ( settings 18 and 19 4:2:2). The original file is lagarith pal i know that i will loose quality but visually they are almost identical and i need space on hdd to keep all the files for now ( don't suggest me to buy hdd or such). The intention is to have the file as close to the original as possible ( without noise reduction or any processing prior to compressing for further processing. Ideally i will keep the lagarith or huffyuv file on the hdd but as i sad i don't have too much space and dv and mjpeg compress almost the same ( on 18 mjpeg even better more then 3 times less than lagarith file ). By the way the files are already processed ( filters noise reduction and so on ) to DVD ( 60-80 min per dvd) and mkv's ( very high bitrate and full d1 res) but i want to keep less compresed copy and to delete lagariths to open space to more captures that will follow. The reason i chose this 2 codec's is they are very easily editable and ok on quality is also the mjpeg keeps tff and dv is always bff ( so i think is plus). Is this ok or do you guys suggest anything else x264 ? mpeg? or else ?

Thanks

[jmac698]

From http://forum.doom9.org/archive/index.php/t-158420.html
FFV1 from ffdshow
Colorspace: 422P
Coder type: AC
context Model: Large
keyframes distance: 10
Gave a compression of 2.3:1 for VHS material. This was better than 12 other codecs, including lagarith or huffyuv. That's just for the benchmark, don't use those settings! Instead choose 420I. It will give even better compression.

If that's not small enough, I have some reservations about DV PAL (only), because it uses something called 4:1:1 colorspace which means the color is stored only once for every 4 horizontal pixels. This is really quite low. Color resolution of VHS is very low, however I'd still feel better if you used a 4:2:0 codec. mjpeg is ok. For PAL material, 4:2:2 doesn't make sense, the very meaning of Phase Alternating Line is that the same color appears on each pair of lines.

[2Bdecided]

I have some reservations about DV PAL (only), because it uses something called 4:1:1 colorspace

No, that's NTSC DV. PAL DV is 4:2:0 with DV-specific sample locations.

which means the color is stored only once for every 4 horizontal pixels. This is really quite low.

VHS has about 30 lines chroma resolution - about one pixel in 12 has unique colour information. 4:1:1 is still overkill.

No lossy compressed format handles huge amounts of noise well (unless it's very high bitrate, or a free VBR which runs at a high bitrate to keep the noise). But with moderate noise levels, I agree with you: I can't see the difference. Though in theory it could become visible with future processing (especially sharpening). You can try that to see how bad it could get with your sources. Look for visible blockiness. I can't imagine ever sharpening that much - YMMV.

Cheers,
David.

[lordsmurf]

High bitrate MPEG-2 is just as good as MJPEG, and better than DV.

4:1:1 looks acceptable on paper, but it's really not. It looks awful for converted VHS.
Color gets altered, lost, and blurred when DV 4:1:1 is used to convert VHS tapes.
DV was intended to be a shooting format, not a conversion format.

Lossless would be Huffyuv.

[mammo1789]

High bitrate MPEG-2 is just as good as MJPEG, and better than DV.

But is it not mpeg bad for editing and filtering?

No, that's NTSC DV. PAL DV is 4:2:0 with DV-specific sample locations.

Yes I am in the pal land.
In mjpeg configuration I only see 1:1:1 which i assume is 4:4:4 right? and 4:2:2 ( not 4:2:0) which I am using and lowest 4:1:1 as I understand the 4:2:2 is half color information and 4:1:1 is quarter subsampling. isn't 4:2:0 YV12 and 4:2:2 YUV2 sort of speak so i think i preserve better color information with 4:2:2 right?

[jagabo]

4:4:4 = 1:1:1 = 720x576 luma, 720x576 chroma
4:2:2 = 720x576 luma, 360x576 chroma
4:2:0 = 720x576 luma, 360x288 chroma
4:1:1 (NTSC DV): 720x480 luma, 180x480 chroma

[El Heggunte]

High bitrate MPEG-2 is just as good as MJPEG, and better than DV.

But is it not mpeg bad for editing and filtering?

Not if it uses only I-frames.

[lordsmurf]

Even IP and some IPB can be fine.
Short GOPs are not really an issue, and haven't been for years.

Long GOP is the problem.

[jmac698]

Correct, my mistake. Rather embarrassing that NTSC has the crappy color! But, I have to disagree with simple reasoning based on 30 lines. I've tested something similar, resolution of VHS, and although it was officially some number, there's still visible changes but with smaller amounts. Also sampling theory doesn't apply because the VHS signals I see aren't linear.

http://code.google.com/p/avisynthrestoration/wiki/Testpatterns
A 2.53MHz test pattern from VHS

I'll have to do this again in color...

[jagabo]

http://code.google.com/p/avisynthrestoration/wiki/Testpatterns
A 2.53MHz test pattern from VHS

I'll have to do this again in color...

https://forum.videohelp.com/threads/319420-Who-uses-a-DVD-recorder-as-a-line-TBC-and-wh...=1#post1980652
https://forum.videohelp.com/threads/319420-Who-uses-a-DVD-recorder-as-a-line-TBC-and-wh...=1#post1981589

[lordsmurf]

This thread is almost a flashback to a decade ago (2003-2004).

While some things may "look fine" on paper (all the theory, math, etc), it looks like ass on the TV. DV is the ass, when used to convert VHS. And at the end of the day, I watch video on a TV, not with a calculator. DV takes an already-lousy consumer format, and manages to make it worse.

MJPEG codecs have always been rotten compared to lossless or good MPEG-2 codecs. I'd avoid them entirely. In fact, many of them are now discontinued, and are not even compatible with a modern OS.

I could rehash the "Huffyuv is not really lossless" argument, too, but I've not seen it in years. It's been so long that I don't even remember much about it, and my research samples are long gone.

That is all.

[jmac698]

Great stuff jagabo, but there's no usable scale. I was looking for this bit of trivia for a while, but I think VCR manufacturers used to spec their resolution as the -6db point (which is stretching it, as -3db is the usual intercept), where tv lines would be the 1:1 aspect ratio taking into account ITU sizing, in modern terms, would be 710.85:486. Something like that, no time to look it up now.

[jagabo]

Great stuff jagabo, but there's no usable scale.

Scale? It's 720 pixels wide, corresponding to columns of the full D1 cap. The graph is 256 pixels tall representing all possible Y/U/V values. The solid(ish) line at the top and bottom is the full amplitude of the original signal (well, not quite, the DVD output was low pass filtered reducing the amplitude of the signal wave a bit towards the right edge of the frame). -6 dB would be about 1/4 amplitude. You can easily load the image into a paint program and count the pixels. At -6 dB one full wave would be about 28 pixels wide, corresponding to about 50 lines across the entire image, or about 38 lines in video speak.

[mammo1789]

While some things may "look fine" on paper (all the theory, math, etc), it looks like ass on the TV. DV is the ass, when used to convert VHS. And at the end of the day, I watch video on a TV, not with a calculator. DV takes an already-lousy consumer format, and manages to make it worse.

MJPEG codecs have always been rotten compared to lossless or good MPEG-2 codecs. I'd avoid them entirely. In fact, many of them are now discontinued, and are not even compatible with a modern OS.

But i show it previously that there almost none ( visually speaking certainly ) difference between the 3 and i can post 15000 ( 9000 even more) kbs mpeg2 and there is difference visually. I didn't speak about losless because i don't have space for it ( keeping it long )

I can't agree about mjpeg being obsolete even some photo camera's use it today for video and i use it in windows 7 and windows 8 64x ( they seem pretty new os). Further more isn't it h264 even "better" than mpeg2 (speaking efficiency wise). I had chance to compare cameras that capture in h264 and cameras that record on dvd mpeg2 ( i know not too much bitrate but i think h264 had lower ones) ( both capturing sd 720x576) and the first looked way better ( on TV and on PC).

In that regard what do you suggest lS should i use mpeg2 in stead of mjpeg or using h264 and what setting will be the best one ( meaning something around 8-13 gb per hour ) for sd material captured losless from svhs unit ( for archival purposes )

[2Bdecided]

MPEG-2 Main Profile @ Main Level is by far the most commonly deployed codec (with the additional 9.8MBps bitrate limit for DVD), and the maximum bitrate this allows (beyond DVD) is 15Mbps.

I suggest people who think I-frame only MPEG-2 at 15Mbps looks better than I-frame only DV at 25Mbps are using a broken DV codec. While 4:1:1 (irrelevant to those of us working with PAL) will degrade a pristine source, if it looks awful (especially with VHS) then probably the resampling is broken in your DV codec too.

Using IP or IPB frames (even in a short GOP) implies prediction. Prediction fails with lots of noise - so on the very sources where DV starts to fall down, short block MPEG-2 will fall down. In some other situations, it will be superior - but given that DV is already good enough, this isn't a great advantage.

Of course, I-frame only MPEG-2 at 50Mbps is great. Broadcasters (e.g. the BBC) use this as an SD playout master all the time. But support in consumer equipment is patchy at best, which rules it out as an "archive" format IMO because you may struggle to decode it in years to come. If you have an open source codec that handles this reliably (does ffmpeg?) and you archive this with your video, then you're probably safer - especially if you're confident re-compiling things for other platforms - so this becomes a great option in this case.

I still think it's overkill for VHS.

Cheers,
David.

[jagabo]

The biggest problem with NTSC DV 4:1:1 and VHS caps is how the decoder upsamples the chroma. For display each of the chroma values has to be spread across four luma samples. Some decoders simply duplicate the chroma samples, some will interpolate. Those that duplicate can leave you with visible vertical stripes.

https://forum.videohelp.com/threads/289684-DV-capture-quality-sucks?p=1758114&viewfull=1#post1758114

DV (as well as all DCT based codecs like MPEG 2 and MJPEG) will also introduce DCT ringing artifacts. This is mostly a problem with very sharp video -- which VHS isn't.

https://forum.videohelp.com/threads/294144-Viewing-tests-and-sample-files?p=1792973&vie...=1#post1792973

Otherwise, the Y, U, and V samples have the same range and meaning (ITU rec.601) as they do in HuffYUV or any other YUV codec.

https://forum.videohelp.com/threads/294144-Viewing-tests-and-sample-files?p=1792760&vie...=1#post1792760

[2Bdecided]

Great comparisons jagabo. I especially like that test image.

Some decoders simply duplicate the chroma samples, some will interpolate. Those that duplicate can leave you with visible vertical stripes.

Exactly. Even if you're stuck with this fault for some reason, it's trivial to fix without losing anything else.

Cheers,
David.