Downscaling interlaced video, my results.

farss schrieb am 28.04.2011 um 05:10 Uhr

I've tried applying some science to this. Here's what I tried just to be sure of the very basics in Vegas regarding the de-interlacing method.
My assumptions / theory:

1) Using Interpolate as the de-interlacing method gives the lowest resolution but no artifacts.
2) Using Blend gives better resolution but some artifacting on edges in motion.
3) None is disaster and will produce horrid artifacts.

All my tests were done in 50i however I can think of no reason why anything would change doing this in 60i. Also all these tests were done in V9.0e. As far as I know nothing has changed in V10.

I took the ISO res chart that A, Grant had kindly converted to a hi res png file and dropped that into a 1080i50 project. I zoomed in on it a bit although how much I should have is a bit of a mystery as the chart's 16:9 markers and Vegas don't agree. I'm inclined to trust Vegas and the purpose of the test was only to compare various methods anyway.
At Best / Full with Scaling off the best I can read is around 830 lines. That sits very nicely with what Musicvid mentioned some time ago, there is a limit of 83% anyway. So far so good. I added a bouncing ball using gen media etc so as to have some motion to check the interlacing.
Then I rendered this out as 1080i50 to the MXF 422 codec and bought that back into a new project matched to that. Looked pretty good, can still read the same resolution from the trumpets.
Next step was to render this out three times to SD 50i using the Sony 8 bit YUV codec. Once using de-interlace = None, Blend and Interpolate and then compared the results.

Blend and Interpolate produced EXACTLY the same result. That is alarming but what is more alarming is the res was only 230 lines! None produced over 400 lines resolution but the wierd thing here was there was no sign of interlaced artifacts (dog's teeth) at all on the bouncing ball and my understanding is there damn well should have been.

So as a good scientific experiment goes this one was an outstanding success, all my assumptions have been proved wrong. I shall of course test this with real world footage. This may raise a number of other questions, did something get changed a few years ago in how Vegas does things, have we been making certain asumptions without testing them, am I in an alternate universe.

Bob.

Zurück zum Post

Kommentare

craftech schrieb am 29.04.2011 um 12:40 Uhr

So going forward my plan is to find our printed 4K resolution chart (we moved) and park that in front of the best HD camera I can borrow (PMW 350) hook that up to a Nanoflash, monitor that on a full HD monitor and record the results at 25p, 30p, 50i and 60i. Sharing that around will be a tad more difficult than a Veg file and a png of a res chart but we only need 1 second of each format so hopefully Dropbox to the rescue.
================
Yes, but I don't know if you saw my EDIT before you posted. It doesn't seem to me that the res chart comparison can be done by any more than a single person.

John

farss schrieb am 29.04.2011 um 13:26 Uhr

"Yes, but I don't know if you saw my EDIT before you posted. It doesn't seem to me that the res chart comparison can be done by any more than a single person. "

No I didn't but I have now.
That's certainly a valid point and MTF also comes into that.
On the other hand an error of +/- 20% or even more should be more than acceptable. The error reading a value off a chart is one thing but it's got to be less subjective than looking at a moving image and me saying it looks sharp and you saying it looks soft.
Also we're trying to compare the results of different processes so it's really only the relative values that matter. So if I see a 50% reduction in resolution from a 50% scaling that's good, if you saw a 60% reduction that's pretty much OK too, you need a big change in resolution for it to be noticeable in the real world.

Bob.

craftech schrieb am 29.04.2011 um 15:04 Uhr

"Yes, but I don't know if you saw my EDIT before you posted. It doesn't seem to me that the res chart comparison can be done by any more than a single person. "
==========================
No I didn't but I have now.
===========================
OK, for argument's sake, let's say that 20% is the actual difference in the results that you are getting and John is getting.

John: 375 lines - 20% = 300 lines

Bob: 230 lines + 20% = 276 lines

You live 10,000 miles or more away from John. Unless the two of you are sitting side by side matching exactly the same steps it seems that the difference in results and opinion of what is acceptable may indeed by subjective. Most posts I read regarding HD to SD downconversion including the article I linked by Walter Graff seem to indicate relative dissatisfaction. You say you get better looking video when viewing it in a CRT, I find it absolutely the opposite. My Plasma looks much better displaying the SD downconverted video than my CRT.

Plus, after all, you are using a still image for measurement.

John

farss schrieb am 30.04.2011 um 00:34 Uhr

John,
Why do you think Mr Graff's article has any relevance?

My spin:
He is comparing the downconverted SD from a rather expensive HD camera with the SD from a rather expensive SD camera. The SD camera is already producing SD very close to as good as it gets, there is no wiggle room left for it to get any "better".

Bob.

craftech schrieb am 30.04.2011 um 02:43 Uhr

My spin:
===============
I realize that, but at the end of the article he said he also tried an $80,000 HD CAM and still didn't get any better HD to SD results.

John

John_Cline schrieb am 30.04.2011 um 03:27 Uhr

I am familiar with Walter Graff and his work. Two things: the article that you referenced was written eight years ago in 2003. That is the equivalent of ancient Biblical text in the technology world. Secondly, his career started in 1982 which was fifteen years after mine started but I'm not going to hold his youth and inexperience against him. :-)

Folks, let's face it, even good SD looks like CRAP compared to HD. Most of the people viewing SD aren't going to really care as long as it is lit well, in focus and doesn't look like a cheap webcam. One of my clients sold over $2 million worth of DVDs that I produced for them last year alone and not a single person has complained about them looking the least bit sub-standard. In fact, they got a few comments about how good they looked.

I'm not at all saying the WE shouldn't care and I'm as guilty as anyone when it comes to obsessing over some "flaw" that only I am likely to see. When it comes to making SD from HD, Vegas works perfectly well to my trained eye. Like I said earlier, if you want downconverted HD to look like over-sharpened SD, add the sharpen filter.

ushere schrieb am 30.04.2011 um 04:04 Uhr

i have to agree with john on this - from all my (limited) tests, vegas does as good a job of hd to sd as anything else.

this is not to say there aren't subtle differences in various workflows, but when your audience is used to viewing web/phone cam, youtube and badly transmitted hd and sd, whatever i (we) produce looks pretty bloody good.

and, if your audience is more concerned with picture quality, then the content is certainly lacking ;-)

farss schrieb am 30.04.2011 um 09:41 Uhr

"I realize that, but at the end of the article he said he also tried an $80,000 HD CAM and still didn't get any better HD to SD results."

And why would he?

Assuming the SD camera was something like the DVW 709 digi betacam camera it is impossible to get better looking SD.
SD NTSC is 720x480 pixels. Because it is interlaced there's a limit to the VRES of around 80% which says VRES is limited to 380 lines.
Now if any camera / lens can resolve HRES = 720 and VRES = 380, that's it, you simply cannot get a better result by putting more pixels in the imager or a better lens in front of it.
The system is the limit, not the camera. Graff could shoot the same scene on 65mm, scan it at 8K and make a DB tape and play that and the resolution will not improve, period. The 65mm film will give you more latitude, the chroma sampling will be better so if the image is heavily graded it will hold up better and the SD will look better but that's not because it has more resolution compared to what was shot on the SD camera.

Put simply the best you can hope for from downconverting HD to SD is SD as good as the best SD cameras can record. My experience watching OTA is that there has been a very slight improvement in PQ as the broadcasters have gone from SD to HD cameras, there's also been an increase in the amount of line twitter.

Bob.

NickHope schrieb am 30.04.2011 um 10:06 Uhr

If anyone is interested, I did some interlaced HD to DVD test renders and made them into a ready-to-burn DVD to download. See this thread. Thanks

amendegw schrieb am 30.04.2011 um 10:38 Uhr

John_Cline says: "Guys, seriously, Vegas rescales HD interlaced to SD interlaced just fine. I'm amazed at how often this comes up mainly because there is no problem that I can see. SCS can't make Vegas scale any better because it's already the best it can be."

ushere says: "i have to agree with john on this - from all my (limited) tests, vegas does as good a job of hd to sd as anything else."I hear ya guys - and I certainly respect your expertise in this area as far greater than mine, but my old eyes tell me differently.

I've rendered three versions of the troublesome "Hula Dancer" clip. Downloadable here: Hula-Renders.zip

1) Proj-A.mpg is a straight render/resize from Vegas 10.0c using the MainConcept encoder - project properties are "match meda properties", reset De-Interlace method to "Interpolate" and rendering quality="best".
2) Proj-B-Lagarith.mpg is the best Vegas-only render that I could produce (to my eyes) using a complicated 2 project process - documenteded here: HD to SD Challenge It's not perfect - still displays some moire - but I think it looks a lot better than the direct Vegas resize in 1) .
3) Proj-A-HandBrake.mp4 really doesn't apply, but I threw it in just for comparison purposes - it's a 720x480 progressive HandBrake transcode directly from the AVCHD 1080i mts source clip. Note: you must manually set your player to widescreen.

...Jerry

craftech schrieb am 30.04.2011 um 11:25 Uhr

I have an idea that may demonstrate how subjectivity plays into this discussion.

Go to the thread Nick linked above and download and burn the SD DVD he is providing that contains various render methods of HD clips some of us sent him.

Nick asks:

"Please burn the 2 folders to a DVD and let me know any reactions/preferences you have to it, and please tell us what gear you watched it on (CRT, LCD, DVD software etc.)."

Thanks Nick,

John

farss schrieb am 30.04.2011 um 11:46 Uhr

1) Looks OK given that it's NTSC and the camera used
2) Looks a smidge better in the resolution department but the aliasing is worse in some respects. The high level of HF detail is breaking through into the chroma part of the signal and causing coloured rainbows to dance around, not good.
3) I've never seen anything so bad. I really want to analyse that because it's doing some really wierd s**e to the image, wow, just wow.

All of that was from playing it out of Vegas though a D>A converter and composite video to a CRT monitor which sadly enough is still how a lot of the public watch DVDs.

Bob.

dxdy schrieb am 30.04.2011 um 11:58 Uhr

As I understand it, one of the major obstacles to "high quality" downsizing from HD to SD is numerical...neither 1080 or 720 are evenly divisible by 480.

I wonder if you would get a "better" DVD shooting a scene zoomed out on a HD camera so that the area of interest was only using 720 x 480 of its pixels?

It seems to me this would be tge case since HD cameras supposedly have better lenses than SD cameras in a similar class, and HD sensors are better (not just bigger) than SD sensors of a similar class.

Shoot the scene, zoomed out so the area of interest is half the height of the frame. In Vegas, use Pan/Crop to crop to 720 x 480.

Has anyone tried this?

amendegw schrieb am 30.04.2011 um 12:44 Uhr

"1) Looks OK given that it's NTSC and the camera usedWow! Do we see things differently. I must state that I'm viewing the clips on my LCD laptop.

I'd like to hear others opinions - Is clip 1) Proj-A.mpg "Ok" and clip 3) Proj-A-Handbrake.mp4: horrible - "never seen anything so bad"?

If so - I'm terribly off base and this will be the last I will say on this subject (with the possible exception of responding to Nick_Hope's DVD test - probably next week).

...Jerry

Edit:
Screenprint from Proj-a.mpg:

Screenprint from Proj-a-HandBrake.mp4:

farss schrieb am 30.04.2011 um 13:03 Uhr

"I must state that I'm viewing the clips on my LCD laptop"

Well if I look at the results on an LCD screen the results are quite different for me too, that's almost a given. If you are not checking interlaced video on a device that can correctly display interlaced video i.e. a CRT monitor or one of the very expensive LCD monitors you can completely and utterly lead yourself up the garden path.

Bob.

NickHope schrieb am 01.05.2011 um 04:20 Uhr

john_dennis schrieb am 01.05.2011 um 05:38 Uhr

Did that and I do see a huge difference. The difference is not as pronounced on tests 2, 3 and 3, however.

farss schrieb am 01.05.2011 um 06:44 Uhr

If you burn my project to a DVD and pause the hula dancer clip during playback of the first test, you should see the huge difference in moire between the playing image and the paused image in the first test. "

The reason that happens is because most DVD players and professional video gear when in Pause drop one field and synthesis a field from the other one so what you see is only one field.

Bob.

John_Cline schrieb am 01.05.2011 um 10:23 Uhr

For what it's worth, all of the subjective opinions that I have stated regarding downscaling interlaced HD to interlaced SD have been reached using both 4x3 and 16x9 calibrated CRT broadcast monitors using a high-end DVD player using its component video output. (The 4x3 monitor has a 16x9 mode that reduces the raster to 16x9 while using all 480 lines instead of a DVD player on a typical 4x3 TV which reduces it to 360 lines.) LCD monitors and particularly software players on computers introduce too many variables and too much processing into the equation. Also, using a DVD player that upsamples to 720p or 1080p renders any evaluation invalid.

I've never had any control over what the viewer is using for playback and what sort of processing it might be doing, but I am reasonably certain of what I'm seeing because I'm viewing it on interlaced CRT monitors at the video's native resolution. Otherwise, you have no way of knowing if it's the file or something downstream that's affecting the image. I know what's leaving my suite but I have no control over what happens to it after that.

I'll download Nick Hope's test, burn it to a DVD and see what it looks like on Monday.