Best process for converting lossless avi to mp4?

[pinkphantom]

Hi there,

I've been reading through forum posts for a while now trying to figure out the best way to do this, but I'm feeling overwhelmed because so much of the advice on here seems to conflict with other people's advice. So, I decided to just make my own post. At the very least, writing all this out might help me organize my own thoughts.

I'm in the process of transferring VHS tapes, VHS-C tapes, and Hi-8 tapes to digital. I'm doing all my Hi-8 tapes first, so that's what this post is dealing with. I'm using the Hauppauge USB-Live 2, with an S-Video cable, to capture and I'm using AmaRecTV with Lagarith to record to lossless avi files. Everything's good up to this point.

Now, my goal is to take these source recordings, deinterlace them, denoise them (with Neat Video for VirtualDub), crop out the scan lines at the bottom (VirtualDub seems to be fine for this), improve the audio quality (if possible), and compress them to mp4 files so they can be easily shared over the internet. I would like to get some of them on DVDs later too, but that's secondary to the mp4 files.

Here are my questions:
1. What order should these things be done in? Should I be saving a new version of the file after every processing stage? (As opposed to telling VirtualDub to deinterlace and denoise at the same time, for instance)
2. What's the best program for deinterlacing? I saw someone say VirtualDub's deinterlacing isn't the best, but I don't see anything wrong with it.
3. What's the best way to improve audio quality? Someone recommended Audacity, but I find it hard to work with.
4. What's the best program and settings for compression and conversion to mp4? I've tried this so many different ways and it always comes out sort of disappointing. I feel lost when it comes to selecting bitrates and such. I'll take somewhat large file sizes if it means decent quality.

[rgr]

First, check if the image is in the 16-255 or 16-235 color range (DaVinci Resolve)
My Toshiba VHS sends analog in the range of 16-255, while Panasonic in the range of 16-235. Such a curiosity.

1. Deinterlace (yadif is enough).
2. Change the resolution to 720p or 1080p, because Neat is better at removing noise and sharpening.
3. Crop the image not only from the bottom, but also the sides and top.
4. Save to mp4.

And I would like to add that the sound from Panasonic is simply awesome compared to Toshiba. So I don't have to do anything with the sound anymore

[lollo]

1. What order should these things be done in?

mask (do not crop) the head switching noise, deinterlace, denoise (,sharpen).

Upscale is another story, follow the recent threads in the restoration subforum.

The video captured by the Hauppauge USB-Live 2 will always be in YUV 16-254 range, so adjust levels to 16-235 if working with additional filters requiring RGB colorspace.

Should I be saving a new version of the file after every processing stage? (As opposed to telling VirtualDub to deinterlace and denoise at the same time, for instance)

Not necessary.

2. What's the best program for deinterlacing? I saw someone say VirtualDub's deinterlacing isn't the best, but I don't see anything wrong with it.

QTGMC, but requires AviSynth or VapourSynth, so you should change your workflow, which I suggest anyhow. In AviSynth / VapourSynth you will also find many denoisers with similar/better preformances than NeatVideo. And excellent upscalers if required.

3. What's the best way to improve audio quality? Someone recommended Audacity, but I find it hard to work with.

Except for fixing specific issues (clicks, hissing noise), in general the audio of a captured VHS should be ok and no restoration is required.

4. What's the best program and settings for compression and conversion to mp4?

just run a simple command line, for example:

ffmpeg.exe -i <input file> -c:v libx264 -crf 17 -preset slow -aspect 4:3 -c:a aac -b:a 128k <output file>

[rgr]

4. What's the best program and settings for compression and conversion to mp4?

just run a simple command line, for example:

ffmpeg.exe -i <input file> -c:v libx264 -crf 17 -preset slow -aspect 4:3 -c:a aac -b:a 128k <output file>

After cropping, the display aspect ratio does not necessarily have to be 4:3, so I would set the pixel aspect ratio to 16/15 (for PAL) rather than display aspect ratio.

[Sharc]

I've been reading through forum posts for a while now trying to figure out the best way to do this, but I'm feeling overwhelmed because so much of the advice on here seems to conflict with other people's advice....

... and there's a good chance that the controversy will continue when you ask for "whats the best .....", I am afraid
It depends on the tools you have, on the sources, on your knowledge and on subjective preferences.
A good investment of your time would be to learn avisynth basics (assuming you are not yet familiar with it). The recommendations you collect from members here will become more fruitful then.
Or you may want to upload a short sample of your capture to receive specific advice, e.g. for a script.

[lollo]

After cropping, the display aspect ratio does not necessarily have to be 4:3, so I would set the pixel aspect ratio to 16/15 (for PAL) rather than display aspect ratio.

That's why I said "mask do not crop". The PAL capture of the Hauppauge is 720x576, DAR=4/3, SAR=5/4 and PAR=16/15 already (NTSC 720x480).

The avi container has no flag for proper siplaying so the 4:3 DAR must be specified when compressing to mp4.

To be more accurate, the 720x576 frame after filtering should be cropped to 704(702)x576, because ITU-R BT.601-4 (formerly "CCIR-601" or "Rec.601") says so

Cropping the head switching noise in the y direction without compensating may generate distortion in the picture.

[Barrythecrab]

Take a look at Selur’s Hybrid, everything is easier in Hybrid, and it writes the code for you.

[rgr]

After cropping, the display aspect ratio does not necessarily have to be 4:3, so I would set the pixel aspect ratio to 16/15 (for PAL) rather than display aspect ratio.

That's why I said "mask do not crop". The PAL capture of the Hauppauge is 720x576, DAR=4/3, SAR=5/4 and PAR=16/15 already (NTSC 720x480).

The frame will certainly be cropped more than 704x576 and it probably won't keep the 4:3 DAR (which is not necessary anyway).
So why bother? Simply set PAR 16/15 for compression :) If the tool is VirtualDub, it even has to enter it in the encoder settings.

PS. I think Hauppauge doesn't always produce the 16-255 range - rather it exactly duplicates the VCR range. And this one could be 16-235.

[Sharc]

The frame will certainly be cropped more than 704x576 and it probably won't keep the 4:3 DAR (which is not necessary anyway).
So why bother? Simply set PAR 16/15 for compression ....

Why bother? As the OP is capturing VHS and other analog sources the PAR of his capture is very most likely 12:11 rather than 16:15 for PAL. For NTSC the PAR would be 10:11.
The OP didn't even tell us if it is for PAL or NTSC, but as he is from Canada chances are that it is for NTSC.
Cropping to 704 width (and masking the head switching crud) and then encoding for DAR 4:3 would work in both cases.

[lollo]

Why bother? As the OP is capturing VHS and other analog sources...

Yes Sharc. I do not understand why rgr wants to complicate this stuff talking about PAR and vertical cropping. No need at all. The VHS capture is a 4:3 aspect ratio SD video, that's all to care about in this case.

Cropping to 704 width (and masking the head switching crud) and then encoding for DAR 4:3 would work in both cases.

That's what I wrote. If the OP does not care about the 2% error, after the head switching noise masking, he can avoid the crop to 704xXXX and encode the 720xXXX frame. What is important is that he specifies (add) the DAR flag in the final encoded file, because the captured avi file does not have it.

PS. I think Hauppauge doesn't always produce the 16-255 range - rather it exactly duplicates the VCR range. And this one could be 16-235.

The VCR has no range. The card is digitizing an analog signal, and its video decoder does this creating a Y-range 16-254. Whatever the input is.

If the analog luminance at the input has no "data" in the whites (dark scenes), the digitized video will behaves a Y-range 16-XXX, where XXX is the correspondent higher "white" peak in the signal, i.e. 16-219, just to write a mumber.
edit: If the analog luminance at the input has "data" in the deep blacks, Y<16, and the deep whites Y=255, they will be crushed/clipped (not captured).

[rgr]

Why bother? As the OP is capturing VHS and other analog sources...

Yes Sharc. I do not understand why rgr wants to complicate this stuff talking about PAR and vertical cropping. No need at all. The VHS capture is a 4:3 aspect ratio SD video, that's all to care about in this case.

As we are digitizing an image that will never be displayed on a 4:3 screen again, there is no need to keep this aspect ratio.
You just have to crop the parts that do not contain the correct image and save with the correct PAR. I always cut the material to around 560 lines.

PS. I think Hauppauge doesn't always produce the 16-255 range - rather it exactly duplicates the VCR range. And this one could be 16-235.

The VCR has no range. The card is digitizing an analog signal, and its video decoder does this creating a Y-range 16-254. Whatever the input is.

If the analog luminance at the input has no "data" in the whites (dark scenes), the digitized video will behaves a Y-range 16-XXX, where XXX is the correspondent higher "white" peak in the signal, i.e. 16-219, just to write a mumber.
edit: If the analog luminance at the input has "data" in the deep blacks, Y<16, and the deep whites Y=255, they will be crushed/clipped (not captured).

Well, my Panasonic recorder exactly matches the output levels from the VCR. And I get the range 16-235 (from Panasonic VCR) or 16-255 from (from Toshiba VCR). And it is visible.
So by digitalizing the video with Hauptpage, we have a good chance that we will simply destroy our deep whites.

[lollo]

As we are digitizing an image that will never be displayed on a 4:3 screen again, there is no need to keep this aspect ratio.

The other way around: I keep the 4:3 aspect ratio, that is how the image should be watched, whatever display I use.

...the output levels from the VCR...

The output level of a VCR is an analog signal, there is not a concept of 16-235 or 16-255 range. The Toshiba VCR sends/provides a higher brigth level. Your Panasonic recorder is able to digitize in the 16-255 range, so it maps in that range. Beeing the Y level lower in the Panasonic VCR, it will translate in a limited range; the fact that you end up with the exact number 235 (TV-range max Y level) I suspect is a coincidence.

So by digitalizing the video with Hauptpage, we have a good chance that we will simply destroy our deep whites.

The opposite. USB-Live 2 capture range is 16-254, then there is no clipping of the whites. It crushes the deep blacks of the analog domain, i.e. the equivalent "digital" blacks below level 16 (Y<16).

[rgr]

As we are digitizing an image that will never be displayed on a 4:3 screen again, there is no need to keep this aspect ratio.

The other way around: I keep the 4:3 aspect ratio, that is how the image should be watched, whatever display I use.

Of course you can. But why watch an image with distortion on all sides? In general, it's better to cut it out, and the actual image will be a bit bigger and will be scaled anyway. (Although recently I am inclined to the thesis that it is better to scale with good filters at least at 720p).

...the output levels from the VCR...

The output level of a VCR is an analog signal, there is not a concept of 16-235 or 16-255 range. The Toshiba VCR sends/provides a higher brigth level. Your Panasonic recorder is able to digitize in the 16-255 range, so it maps in that range. Beeing the Y level lower in the Panasonic VCR, it will translate in a limited range; the fact that you end up with the exact number 235 (TV-range max Y level) I suspect is a coincidence.

I think that it is more like Toshiba's (and Samsung's method too, because they are rather clones) for a better image. It is indeed nicer, sharper, but so what if overbrightning is clearly visible without processing.

So by digitalizing the video with Hauptpage, we have a good chance that we will simply destroy our deep whites.

The opposite. USB-Live 2 capture range is 16-254, then there is no clipping of the whites. It crushes the deep blacks of the analog domain, i.e. the equivalent "digital" blacks below level 16 (Y<16).

Actually. But I have not seen a VCR that generates Y levels in the range of 0-15.

[lollo]

Of course you can. But why watch an image with distortion on all sides?

There is no distorsion at all. You probably meant black borders on each side. Because if you want to remove all the black borders you have to resize (lossy operation), and do it properly with the right proportions case by case, to do not introduce distorsion. In PAL domain my analog captures have "active pixels" inside the 720x576 captured frame like: 698x544, 688x560, 690x560, 690x562 688x564, 704x552, 688x570, 694x566, 694x570, ...

But I have not seen a VCR that generates Y levels in the range of 0-15

All of them; you are just not able to capture the 0-15 range, which, again, does not exist in the analog domain. You should refer to "analog blacks corresponding to 0-16 once digitized"

[rgr]

Of course you can. But why watch an image with distortion on all sides?

There is no distorsion at all. You probably meant black borders on each side. Because if you want to remove all the black borders you have to resize (lossy operation),

What resize? For what? If you get an picture, e.g. 690x550, then you record 690x550.

But I have not seen a VCR that generates Y levels in the range of 0-15

All of them; you are just not able to capture the 0-15 range, which, again, does not exist in the analog domain. You should refer to "analog blacks corresponding to 0-16 once digitized"

The black color in the VHS signal I can see only on the edges outside the active area.