You can copy the link for the picture to the clipboard ...

The other way to figure it out is that the second is obviously better.

The whole point of the post is to demonstrate why scaling in Vegas does not live up to its "pro" designation ...

[edit after Grazie's response] ... when it comes to moiré.

This is why I like TMPGEnc, it uses Lanczos3. Frameserve from Vegas to TMPGEnc for MPG2 renders.

You can just copy the image location to the clipboard and the answer will be revealed: bicubic is the first one. I didn't know that! - And NOW I do!


The other way to figure it out is that the second is obviously better, Not to me. But now you have confirmed this, good. Which of the sections should I be looking at to say "That's right!"


. . and the whole point of the post is to demonstrate why scaling in Vegas does not live up to its "pro" designation.I have no doubt it was, and I see now just what is meant. But it's so good to have these things spelt out.

I suppose a simple TITLE to each would have had been neat.

Grazie

As a hobbyist/amateur, I'd also like to be educated a bit about where to spot the differences and what to look for. (I'm not being sarcastic -- I'd love to hear more!)

The 4, 3, 2, 1 test patterns in the bottom chart show much less moire patterns, with #1 resolving to a nice, even gray rather than the "interference fringes" that result in the top chart.

On the other hand, the green text and green lines all look decidedly worse to me on the bottom chart, as well as the pretty color bars near the "100%" legend (in fact all the color bar boundaries) looking much worse on the bottom.

I would guess, though, that with video in motion the moire issues would be much more distracting than the text/line/color bar issues. But further illumination here would be appreciated!

+Tim

I looked at the Green Lines and the GREY squares too . . and I confused myself.

G

The problem with moiré is that the effect depends not only on the source and how it is scaled, but also on what resolution your monitor is set to when you view the resulting images. This is because the monitor introduces another level of possible moiré effect. You can see this if you download the two images and then view them at different magnifications (i.e., zoom into the image). The bicubic provides a much cleaner scaling of the resolution chart, and you can resolve more lines. However, at certain magnifications, this creates the "beat" patterns that are characteristic of the moiré effect.

We went through all this years ago with that amazingly long thread about re-sizing.

Interlaced HD to DVD AGAIN - some test renders

The key image that was used in that case was the picture of a hula dancer doll, placed on top of a fine-patterned B&W checkered cloth. That clip can still be found here:

HD to SD Challenge

Like the resolution chart, that hula doll video is what we engineers call a "pathological case," meaning that it is not particularly representative of what happens in 99% of all video. My conclusion from that long discussion is that the approaches which did the best job eliminating the moiré were not always the methods which produced the best looking video when non-patholgical test clips were used.

However, as already pointed out, the one constant in all of these discussions is that Vegas does a poor job with the deinterlacing job that must be done when re-sizing interlaced video. Thus, as already stated, when resizing a source that is progressive, Vegas can do a fine job as long as you use "Best" when rendering.

So, I too would like to get a few seconds of a source clip from the OP, because I think there is something else going on in this particular case.

Here, the green lines and letters are best in bicubic, and the patterns are better in lanczos... but no one uses charts for telling stories :)

My two cents.

Erni

Unfortunately, Vegas cannot use Lanczos natively. It would ostensibly represent a COI with their commercial licensors. Nor any of the really good deinterlacers for that matter.

In real-world testing where moire is not a big issue, I actually found the Lanczos-sized material a little softer, besides taking longer. That's just an opinion ;?)

Since this discussion has been going on for at least four years in this forum, it surprises me that SCS hasn't implemented Lanczos as an option. Is it that hard to do?

/jerry

Jerry, we cross-posted.
Lanczos is open source. I would love to see you, or someone write a plugin.

It's time for us to take to the ramparts and storm the Bastille!

"Let us have our Lanczos!"

......

>Unfortunately, Vegas cannot use Lanczos natively.

I found this interesting post on the Adobe blog which says Premiere uses Lanczos when using GPU for scaling. This is because the massive parallelism of GPU is necessary to get satisfactory speed with Lanczos.

When Premiere Pro is just using the CPU for the processing of scaling operations, it uses the following scaling methods:

Edit: But it says "sampled with bicubic" - not sure what that means and how it differs from what Vegas is doing.

/jerry

That's why I switched to Adobe CS6 for all of my HD to DVD conversions. It just looks better. I used to use TMPEG which also did a good job.

It might be time for anyone who is interested in doing proper testing to get the right tools. The test patter being used here was never designed for testing resolution and aliasing.
Zone plates are the go to tool for this but from my own research the full set of tools including the software to analyse the results is expensive. I did just find something way, way more affordable thanks to a collaboration with Allan Roberts from the BBC and BRP.

Alan Roberts “Signature” range of HD test


Based on a discussion with the SoFo engineers who wrote the code in question what Vegas needs is a low pass filter. There will always be a trade off between aliasing, resolution and render time. An adjustable low pass filter would give the Vegas user the ability to decide how he wants to make that trade off. There's no perfect solution, watch some OTA television and you'll always notice some aliasing and that's out of hardware scalers that cost big time.

Bob.

Those are nifty charts indeed, but I've worked with conventional H-V resolution charts since elementary school. Also, those are a paid solution.

The resolution bars in Belle-Nuit actually measure an aggregate of encoder resolution + chroma subsampling artifacts. a valuable tool when comparing RGB source with YUV renders, actually the supertopic of this discussion.

This, from the Belle-Nuit docs:
"Horizontal and vertical resolution
Better than charts, the best baseline comparison tool is called PSNR. Easy to do in Handbrake, but would require an external utility (or plugin?) in Vegas. Since we are dealing with resolution, SSIM is probably not necessary.

A footnote on lowpass: There is open-source NL-Means algorithm actually in use that is 'way better than linear lowpass or even Median, but predictably, is godawful slow.

Many years back, I put together an archive of some classic monochrome res charts, including USAF 1951, and the original NTSC EIA1956, here:
http://dl.dropboxusercontent.com/u/20519276/Test%20patterns-1.zip

[I]"
Better than charts, the best baseline comparison tool is called PSNR"[/I]

Sure Signal to Noise Ratio is a useful tool but from my limited knowledge it's not going to give you much diagnostic information. A zone plate gives you a swept frequency from DC to beyond the Nyquist limit which is a very good diagnostic tool for both cameras and NLEs.

Here's a good post from Alan Roberts, Zone plates, aliasing, and other things giving a good explanation of the issues and thankfully without all the maths which goes right over my head :(

Bob.

I tried a whole lot of resize algorithms in AviSynth, with one done just in Vegas for comparison.


gave a very interesting clean result, worthy of more investigation.
5. I'll probably continue to use Spline36Resize > vertical-only blur1.0 > sharpen0.75 for my own HD>DVD conversions.

Also, there is a useful comparison of resize kernels here: [url=http://svn.int64.org/viewvc/int64/resamplehq/doc/kernels.html]

Edit: Here is the ResampleHQ version:

I posted some of this in another thread.

I think these discussions fail to account for shooting progressive vs shooting interlaced.
Lots of Panasonic and Canon cams shoot natively in progressive. I have a Canon that does that. But my EX1 is capable of both and I always choose interlaced to shoot all my stage productions.

Why?

Because the light gathering ability of interlaced shooting is far greater than that of progressive.

So comparing workflows and results of the various HD to SD methods should distinguish between the two, no?

Moreover, I get motion artifacts particularly on actor's faces under stage lighting. They are motion artifacts. How is rendering stationary test charts a comparison of the results in real world video?

John

"Because the light gathering ability of interlaced shooting is far greater than that of progressive"

As a one-time politician in Aus. famously demanded when she was confused, "Please explain!".

It has been said that Vegas needs a low pass filter. Blurring is a type of low pass filter, and Vegas has several options. Vegas may not have the optimum filter for the job, but Gaussian blur, say, should be better than nothing.

The sequence should be for interlaced HD to interlaced SD:

Deinterlace to twice frame rate (eg 50i to 50p)
Low pass filter
Reduce resolution (HD to SD)
Re-interlace by discarding unwanted lines (eg 50p to 50i)
Sharpen and add salt to taste

The sequence should be for interlaced HD to interlaced SD:

That is exactly what I have been doing and that is also good way to get to 30p with Vegas only.

I can get similar results straight from the timeline versus HB when I apply an UnsharpMask and Convolution Kernel filter package but it takes quite long to get it rendered and it is even a little more softer then the HB route.