Converting from 720x576 (Mini DV PAL) to 768x576 (square pixels)

[Laskaris]

Hello all!

I have a bunch of MiniDV tapes which I captured as .avi files via Firewire/WinDV. Now, as a first step before further processing, I'd like to deinterlace the videos and get the correct aspect ratio, using the Hybrid software.

Since I'm in Europe, my MiniDV tapes are PAL, so they use a native resolution of 720x576 with rectangular 16:15 pixels. Using Hybrid, I can convert the output to square pixels, which means that, keeping the same number of horizontal lines, I get a resolution of 768x576.

Now for my question: What is the best resize method to use for this task?

Many experts recommend using the NNEDI3 resize method, which is supposedly the best. However, as far as I understand, NNEDI3 can only upscale the video by a factor of 2 and then, if needed, the video is downscaled again to the desired resolution using a different algorithm. So, in my case, the video would be upscaled to (720x576) x 2 = 1440x1152 pixels with NNEDI3, and then it would be downscaled to 768x576 using the specified downsizer (the default in Hybrid seems to be spline64).

So, this makes me wonder: I want to do only a minor upscale to convert the 16:15 pixels to square pixels, going from the native MiniDV resolution of 720x576 to 768x576, but if I use NNEDI3 to do that, the video first gets upscaled by a factor of 2, to a fairly big resolution of 1440x1152, before being downscaled again. Won't that introduce problems, creating artifacts and such?

Would I be better off if I simply use the bicubic spline resize method for upscaling from 720x576 to 768x576?

I'm looking for advice on what the best practice is. Thank you in advance!

[videoAI]

try
clever FFmpeg-GUI
use Lanczos [good for quality/detail]


[Attachment 89546 - Click to enlarge]

[pandy]

Since I'm in Europe, my MiniDV tapes are PAL, so they use a native resolution of 720x576 with rectangular 16:15 pixels. Using Hybrid, I can convert the output to square pixels, which means that, keeping the same number of horizontal lines, I get a resolution of 768x576.

But why? 768x576 is not standard resolution - with correctly signaled aspect ratio there no need to do such conversion.

[dellsam34]

I would decode from DV to basic AVI, de-interlace, resize directly from 720x576 to 1440x1080 and encode for playback/sharing, It's future proof and will be compatible with future displays (4K, 8K ..), all the TV has to do is double the lines.

I use the same process for DV (DV tapes and D8 tapes), Here is a sample.

[Goku73]

The correct process would be to change the resolution from 720x576 to 720x540 and not to 768x576, which is the old anamorphic format now only used on DVDs for old CRT TVs. In fact, to be more precise, since ITU almost certainly has to be activated the ratio changes from 16:15 to 12:11, the correct resolution should be 720x528 with a ratio of approximately 1.37, otherwise you would end up with elongated faces and oval spheres. To remove interlacing, one of the best scripts is QTGMC by avisynth, and as a resizer for these formats, it would be best to use Spline36Resize. I don't know if Hybrid uses avisynth, but it would be best to use a program that does.

[Sharc]

The correct process would be to change the resolution from 720x576 to 720x540 ....

Huh, vertical resizing of interlaced video, really? At least you should deinterlace as the first step if you want to resize vertically.
And yes, Hybrid supports both Avisynth and Vapoursynth.

[Laskaris]

Thank you for your replies.

I did mention it in my original post, but perhaps I should have emphasized it more: I want to do the conversion from 720x576 to 768x576 only as a first step, to get an intermediate for further processing (including, potentially, AI upscaling with a program like Topaz Video AI).

Ultimately, I'd like to upscale the video to 960x720 pixels or even to 1440x1080 if the source material is clean enough to not look too artificial. But first off, I'd like to deinterlace and get the correct aspect ratio. My understanding is that the correct aspect ratio for my MiniDV PAL source material would be 4:3, hence 768x576 pixels and not 720x576. Or am I wrong about that?

The correct process would be to change the resolution from 720x576 to 720x540 and not to 768x576.

But if I change the resolution to 720x540, I lose some of the information of the original video. And since I intend to, ultimately, upscale the video to 720p or ideally 1080p, I need to keep all the info for later processing. So during the deinterlacing/pixel ratio correction, I should upscale to 768x576 instead of downscaling to 720x540. At least that seems to me to be the logical way to do it, to keep all the original data and upscale it instead of downscaling it - or am I misunderstanding something here?

[dellsam34]

Avoid multiple resizing, it will hurt the quality, Do it once after de-interlacing, whether AI or not, and most importantly, work in lossless (or losslessly compressed) AVI, even if you are not gaining anything from decoding 4:2:0 DV into 4:2:2 AVI, the processing is less destructive in this format while working towards your final resolution and codec.

[Laskaris]

Avoid multiple resizing, it will hurt the quality, Do it once after de-interlacing, whether AI or not, and most importantly, work in lossless (or losslessly compressed) AVI, even if you are not gaining anything from decoding 4:2:0 DV into 4:2:2 AVI, the processing is less destructive in this format while working towards your final resolution and codec.

I'm still a beginner at this, so I am unsure about the ideal workflow.

My original plan was to deinterlace with Hybrid and, at the same time, convert the 16:15 pixels to square pixels (hence the upscale from 720x 576 to 768x576).

Then I was going to do color grading with DaVinci Resolve.

Then I wanted to do the upscale to the desired final resolution of 960x720 or 1440x1080. One common advice I keep reading is that the upscaling should be done after the color grading, not before. Is that correct?

So, if multiple resizing is to be avoided, should I only deinterlace the material first, keeping the 720x576 resolution, then do the color grading, then do the resize (with or without AI) to 960x720 or 1440x1080?

[dellsam34]

You can do all the desired restoration steps in a de-interlaced AVI 4:2:2 1440x1080 @ 4:3 aspect ratio, Most software are happy with it as it is the very basic video format and HD resolution paved the way for future screen resolutions in terms of display compatibility unlike any resolution below HD where complex algorithms are needed to upscale.
I personally decode DV using vdub2 into interlaced lossless AVI, Then I de-interlace with QGTMC and resize to 1440x1080, from this I can upload to YT, or encode and do restoration if needed.

[Goku73]

You can do all the desired restoration steps in a de-interlaced AVI 4:2:2 1440x1080 @ 4:3 aspect ratio, Most software are happy with it as it is the very basic video format and HD resolution paved the way for future screen resolutions in terms of display compatibility unlike any resolution below HD where complex algorithms are needed to upscale.
I personally decode DV using vdub2 into interlaced lossless AVI, Then I de-interlace with QGTMC and resize to 1440x1080, from this I can upload to YT, or encode and do restoration if needed.

Exactly, deinterlace with QTGMC, first bring it to 720x540 as I said before, which is the base, and not 768x576 because you are doing unnecessary upscale, and then, if you want, bring it to 1440x1080, which is nothing more than 720x2 and 540x2. But I'm convinced that you'll still have the wrong aspect ratio because I believe that any camera records in 1.37 and not 1.33, so 720x576 12:11 and not 720x576 16:15. You'd just need to do a test to understand this. 90% or more of 4:3 SD (Standard Definition) source footage from DVDs or camcorders transferred to Blu-ray at 1440x1080 all have the wrong aspect ratio. Example:


All Dragon Ball Blu-rays and streaming videos have been upscaled from DVD to 1440x1080, which is wrong, and those who sell them throughout Europe have done so.

[Sharc]

Exactly, deinterlace with QTGMC, first bring it to 720x540 as I said before, which is the base, and not 768x576 because you are doing unnecessary upscale ....

You are proposing unnecessary vertical downscale to 540, actually throwing away original picture details/resolution. Usually a bad idea to unnecessary damage the source in a first step.

[rgr]

I would decode from DV to basic AVI, de-interlace, resize directly from 720x576 to 1440x1080 and encode for playback/sharing, It's future proof and will be compatible with future displays (4K, 8K ..), all the TV has to do is double the lines.

I use the same process for DV (DV tapes and D8 tapes), Here is a sample.

Why? The TV will do a better upscaling in one pass from 720x576 to 4k than in two passes.

[Goku73]

Can you run an avinaptic report? Are you sure you're not already in this situation?

Video track

Codec ID V_MPEG2
Resolution 720 x 576
Display resolution 768 x 576 (pixels)
Frame aspect ratio 5:4 = 1.25
Pixel aspect ratio 16:15 = 1.066667
Display aspect ratio 4:3 = 1.333333
Framerate 25 fps

[Laskaris]

But I'm convinced that you'll still have the wrong aspect ratio because I believe that any camera records in 1.37 and not 1.33, so 720x576 12:11 and not 720x576 16:15. You'd just need to do a test to understand this. 90% or more of 4:3 SD (Standard Definition) source footage from DVDs or camcorders transferred to Blu-ray at 1440x1080 all have the wrong aspect ratio.

The footage was recorded with two DV camcorders, a Panasonic NV-GS17 and a Panasonic NV-GS21. I can't say for sure whether they recorded with a pixel aspect ratio of 16:15 or 12:11. There is no mention of PAR anywhere in the manual.

Exactly, deinterlace with QTGMC, first bring it to 720x540 as I said before, which is the base, and not 768x576 because you are doing unnecessary upscale ....

You are proposing unnecessary vertical downscale to 540, actually throwing away original picture details/resolution. Usually a bad idea to unnecessary damage the source in a first step.

That is my thinking exactly, which is why I am reluctant to follow this particular piece of advice.

[Alwyn]

@Goku73, I can't see why the SAR has anything to do with this. This is DV. The display ratio is 4:3. The file is 720 across and it will be displayed as 768. If the 720 file is resized to 768 square, the visual result will be exactly the same. In my experience with DV files there is no distortion when resizing a 4:3 720x576 file to 768x576 square pixels. Here's a wheel from an unedited PAL DV file that has been resized to 768x576 with VDub2 (720>1440 shows the same):


[Attachment 89562 - Click to enlarge]

As for your DVD>Bluray Anime, unless you know every processing step, who knows where it went astray.

My philosophy? Gettem square and keep 'em square. For resizing I just use VDub2 to go from 720x576 to 768x576 as part of the deinterlacing process ie at the end. If I have no editing at all to do, I'll go to 1440x1080 in VDub, or if I'm editing in my NLE, I'll resize as part of the export process after editing. I don't know what algos they use. I don't go straight to 1440x1080 after double-rate deinterlacing because the AVI files are just too big to be manageable (8 times bigger than source).

In your case, using Resolve, be mindful that it won't accept "normal" AVIs, only DV and ProRes. So using Hybrid, you'll need to export into a format that Resolve will accept, but ideally not back to DV after deinterlacing.

[Sharc]

@Goku73, I can't see why the SAR has anything to do with this. This is DV......

It does, and it seems to depend on the DV camera model. Early models strictly followed the Rec.601 sampling principle (13.5 MHz luma sampling frequency) which has been adopted for the DVCAM (SONY) standard. So the SAR (aka PAR) is ~12/11, and the 4:3 picture is within the ~702 width and maps to 768x576 (PAL) for 4:3. The full 720 picture would actually map to 786x576 =1.36 rather than 1.333... for an undistorted picture, means a circle remains a circle.
Later DV models (especially those which had a setting for 4:3 or 16:9) seem to have violated the original standard and followed the (lazy) DVD practice, so the 720x576 maps to 768x576 for 4:3, aka SAR=16/15. This seems to be your case for your camera model.
We have been discussing this ambiguity some time ago, based on your and other users examples.

My philosophy? Gettem square and keep 'em square.

Nice, provided one does it correctly...... otherwise live with the ~2.3% aspect ratio error.

[dellsam34]

first bring it to 720x540 as I said before,

Stop right there, wrong advise, why would someone downscales? and to a non standard resolution, that's nuts.

[dellsam34]

Why? The TV will do a better upscaling in one pass from 720x576 to 4k than in two passes.

I disagree, My then $3000 OLED TV does a horrible upscaling job of video tape sources, They look much better when de-interlaced and upscaled to HD.

[Laskaris]

@Goku73, I can't see why the SAR has anything to do with this. This is DV......

It does, and it seems to depend on the DV camera model. Early models strictly followed the Rec.601 sampling principle (13.5 MHz luma sampling frequency) which has been adopted for the DVCAM (SONY) standard. So the SAR (aka PAR) is ~12/11, and the 4:3 picture is within the ~702 width and maps to 768x576 (PAL) for 4:3. The full 720 picture would actually map to 786x576 =1.36 rather than 1.333... for an undistorted picture, means a circle remains a circle.
Later DV models (especially those which had a setting for 4:3 or 16:9) seem to have violated the original standard and followed the (lazy) DVD practice, so the 720x576 maps to 768x576 for 4:3, aka SAR=16/15. This seems to be your case for your camera model.
We have been discussing this ambiguity some time ago, based on your and other users examples.

Approximately when was this divide between "early" and "later" DV camcorders? According to info I googled, the Panasonic NV-GS17 I have was produced from 2005. Which sounds about right - I remember buying it for a project I did that year.

It does not have an option to record in 16:9, only in 4:3.

You could also take still pictures with the camera. They have a resolution of 640x480, which is exactly 4:3.

[Sharc]

@Goku73, I can't see why the SAR has anything to do with this. This is DV......

It does, and it seems to depend on the DV camera model. Early models strictly followed the Rec.601 sampling principle (13.5 MHz luma sampling frequency) which has been adopted for the DVCAM (SONY) standard. So the SAR (aka PAR) is ~12/11, and the 4:3 picture is within the ~702 width and maps to 768x576 (PAL) for 4:3. The full 720 picture would actually map to 786x576 =1.36 rather than 1.333... for an undistorted picture, means a circle remains a circle.
Later DV models (especially those which had a setting for 4:3 or 16:9) seem to have violated the original standard and followed the (lazy) DVD practice, so the 720x576 maps to 768x576 for 4:3, aka SAR=16/15. This seems to be your case for your camera model.
We have been discussing this ambiguity some time ago, based on your and other users examples.

Approximately when was this divide between "early" and "later" DV camcorders? According to info I googled, the Panasonic NV-GS17 I have was produced from 2005. Which sounds about right - I remember buying it for a project I did that year.

It does not have an option to record in 16:9, only in 4:3.

You could also take still pictures with the camera. They have a resolution of 640x480, which is exactly 4:3.

I can't give you years and models. If you want to know for your camera take a shot (video, not a still picture) of an exact circlar object like a wheel, clock .... (perpendicular, centered) and do the circle test, or post the clip here so someone may take a look.

Edit: Your camera seems to have a "Cinema" mode which gives a 16:9 active picture with a matte (black top/bottom bars in a 4:3 frame).

[Alwyn]

Early models strictly followed the Rec.601 sampling principle (13.5 MHz luma sampling frequency) which has been adopted for the DVCAM (SONY) standard. So the SAR (aka PAR) is ~12/11, and the 4:3 picture is within the ~702 width and maps to 768x576 (PAL) for 4:3. The full 720 picture would actually map to 786x576 =1.36 rather than 1.333... for an undistorted picture, means a circle remains a circle.

So for early DV camcorders, crop 9 off each side and export as 768x576?

[Sharc]

Early models strictly followed the Rec.601 sampling principle (13.5 MHz luma sampling frequency) which has been adopted for the DVCAM (SONY) standard. So the SAR (aka PAR) is ~12/11, and the 4:3 picture is within the ~702 width and maps to 768x576 (PAL) for 4:3. The full 720 picture would actually map to 786x576 =1.36 rather than 1.333... for an undistorted picture, means a circle remains a circle.

So for early DV camcorders, crop 9 off each side and export as 768x576?

Crop total 18 pixels, mod2 horizontally for YUV color space. So I would crop 8 left and 10 right or vice versa, or 8 left and 8 right giving 704 width which is practically perfectly fine, and export to 768x576. Or if you want to keep the full picture export the 720x576 to 786x576 and accept the DAR of 1.36 - don't squeeze it into exact 4:3 for playback. In x264 encodings set the SAR correctly to avoid further conflicts (12/11 or 16/15 - depending what your source really is).

(Btw Vdub uses "59/54 (DV-PAL)" instead of 12/11 in its pixel aspect ratio "catalog". Both represent very good approximations of the exact value).

Added: Personally I usually don't even resize to square pixels but just crop (and perhaps pad to a standard resolution) as needed to remove any border crap and use the correct SAR for encoding, leaving the conversion including the deinterlacing to the player/TV (assuming the player respects the SAR flagging and has a reasonable deinterlacer). Not a general recommendation, just my "old school" preference .

[Alwyn]

For the exercise, here's a DV file of a wheel (the source of my still image above). I don't remember what camcorder I used (I have a few!).

12:11 or 16:15?

[Sharc]

for the exercise, here's a dv file of a wheel (the source of my still image above). I don't remember what camcorder i used (i have a few!).

12:11 or 16:15?

16:15, imo.

There have been many similar discussions in the past (for PAL and NTSC, in various forums, since 2003) and the conclusion one can draw is something like "it seems to depend on the videocam model. Do the circle test to find out what applies in your case, or live with a possible ~2% aspect ratio error". Consumer DV seems not follow a strict standard in this respect.

[rgr]

In your case, using Resolve, be mindful that it won't accept "normal" AVIs, only DV and ProRes.

Resolve accepts H265 12-bit lossless.

[rgr]

Why? The TV will do a better upscaling in one pass from 720x576 to 4k than in two passes.

I disagree, My then $3000 OLED TV does a horrible upscaling job of video tape sources, They look much better when de-interlaced and upscaled to HD.

But that's a defect of your TV By definition, a single upscaling from 720x576 to FHD or 4k should be of better quality than a double upscaling.

[Sharc]

By definition, a single upscaling from 720x576 to FHD or 4k should be of better quality than a double upscaling.

Not necessarily. For a low resolution source upscaling in 2 steps with intermediate sharpening can give better final results.

[dellsam34]

Why? The TV will do a better upscaling in one pass from 720x576 to 4k than in two passes.

I disagree, My then $3000 OLED TV does a horrible upscaling job of video tape sources, They look much better when de-interlaced and upscaled to HD.

But that's a defect of your TV By definition, a single upscaling from 720x576 to FHD or 4k should be of better quality than a double upscaling.

Nope, I have tried several TVs from Plasma to OLED, I was a videophile and all familiar with expensive DVD players with advanced upscaling capabilities, The Plasma HD did better, but LCD sucked the most, Not sure how cheap LCD did but I assume did even worse. You would think modern technology does better but you would be wrong, Low frame processing in software upscaling is the best I've seen, and this is not just my opinion, a lot of folks here have the same experience.

[lollo]

and this is not just my opinion, a lot of folks here have the same experience.

I confirm, not even my Sony A90J is able to match QTGMC deinterlacing and software upscale, despite all latest technologies integrated in embedded hardware and software.

Back to original questions, resizing to square pixel (768x576 or 720x540) is completely useless and (marginally) degrading the quality.