Best way to upscale DV to HD

FilmingPhotoGuy wrote on 2/8/2012, 3:22 AM

Personally I wouldn't upscale all the way to 1920 x 1080 but only to 1280 x 720. Also DV is 4:3 aspect ratio so changing that to 16:9 is stretching it further and not to mention de-interlacing.

That's what I've experience.

Jay Gladwell wrote on 2/8/2012, 6:39 AM

If you have Premiere Pro, could try Magic Bullet Instant HD.

Chienworks wrote on 2/8/2012, 6:55 AM

Well, 4:3 DV is 4:3. Widescreen DV is about 16.3:9

Laurence wrote on 2/8/2012, 7:24 AM

Boris BCC-7 has a really nice uprezzer. Is the DV widescreen interlaced or progressive? Progressive SD actually uprezzer quite nicely.

larry-peter wrote on 2/8/2012, 11:31 AM

You may want to look at this as well:
http://www.infognition.com/super_resolution_vdf/

I just bought this last week ($39.95) and with just a few tests, find it to be pretty impressive as a VirtualDub filter. Got some quirks and limitations, but seems like a good bang for buck when you see the quality.

Here is their "shootout" comparing several of the other techniques mentioned in this thread:
http://www.infognition.com/articles/video_resize_shootout.html

Larry

wwaag wrote on 2/8/2012, 5:46 PM

Larry wrote:

"You may want to look at this as well:
www.infognition.com/super_resolution_vdf/

I just bought this last week ($39.95) and with just a few tests, find it to be pretty impressive as a VirtualDub filter. Got some quirks and limitations, but seems like a good bang for buck when you see the quality."

I'm curious about your workflow. I have the standalone version and on a recent project found absolutely no improvement. My source material was DV (actually Hi8 converted to DV). I first deinterlaced using QTGMC and then did some denoising using one of JohnMeyer's avisynth scripts and saved to an AVI using Cineform. The resulting AVI file was 720x480 at 60P (which actually looked pretty good). I ran that file through infognition's "super-resolution" to produce a 1440x960 file using the highest quality setting. I then created a Vegas project (1280x720) at 60P and imported both files (original plus super-res) to the timeline as different tracks. I then took numerous snapshots (Best-Full) along the timeline of the original and super-res images, imported them into Photoshop, and could not see any difference, regardless of zoom. I must admit I was a bit disappointed.
Perhaps, I've done something wrong or perhaps the initial denoising accounted for the lack of any improvement.

wwaag

larry-peter wrote on 2/8/2012, 9:27 PM

I haven't used the standalone version, but I'll try to get my test shots posted when I get back to my office on Friday. My footage was 24p 720X480 from a DVX100B, so no interlacing was done.. I just ran it through 2 stacked instances of superres in Virtual Dub to get a 4X size. When I put the original in a 1080p Vegas project along with the uprezzed clip, I could read text on a calendar on the background wall, see wood grain details in the paneling and skin details in the superres clip that were gone in the original. The difference was dramatic viewed on a 42" monitor.

Edit: Not knowing that denoising script, it's possible that you're also removing some fine detail that superres would normally be gathering over a series of frames to provide the additional resolution. I'd give it a try without denoising and see what you think.

PeterDuke wrote on 2/8/2012, 10:01 PM

wwaag

I am interested to know what happens if you leave out the de-noising step.

UlfLaursen wrote on 2/8/2012, 11:10 PM

Thanks guys - I'll try these things out :)

/Ulf

riredale wrote on 2/9/2012, 3:11 PM

"Well, 4:3 DV is 4:3."

I remember from my early days on this forum (back when Vegas3 was current) that we had discussions about DV and the fact that it is NOT 4:3.

A 720x480 pixel map with pixel aspect ratios of 0.909 gives a net image aspect ratio of 4.09 : 3. I discovered this when I inserted a 4:3 photo on the timeline and wondered why there was a bit of black on both sides of the preview window.

wwaag wrote on 2/9/2012, 3:54 PM

This morning, I redid a 5 sec clip from a recent project. The raw footage is DV from a Hi8 source. I created an HD-720-60P project in Vegas and then took snapshots of the same frame. I opened each snapshot in Photoshop, zoomed to 300% and did a screen-save using Screenhunter 6.0. To begin, here is the entire screen image not zoomed with the original DV footage frame. (Incidentally, the little girl is my daughter in 1992).

Next, the original DV zoomed to 300%

Next, deinterlaced using QTGMC.

Next, deinterlaced plus super res. Again, rendered to 1440x960 at best quality.

I then rendered the 5 sec clip to an MPG using HD-720-60P, and redid the snapshots thinking the preview might be different from rendered file.

Here is the original DV footage.

Deinterlaced

Deinterlaced plus super-res

Finally, I applied denoising using a Johnmeyer script.

Then, denoising plus super res.

I didn't render these out since I really didn't see any difference between the preview images and rendered images shown previously.

To me, there is a marked improvement from DV to De-interlaced to De-interlaced plus Denoising. However, I couldn't see any real improvement by upscaling using super-res. I did notice a small color shift which could be seen on the Scopes when applying super-res upscaling and can be seen in the above images. Perhaps upscaling another time might improve the outcome, but given the lengthy processing time, I'm satisfied with just the deinterlacing using QTGMC and denoising.

wwaag