AI turns SD into HD: The Future is Now at Advent Digitizing

Standard Definition Video has been around since the beginning of TV, some 70 years ago. It’s noted by the 4:3 “squarish” aspect ratio. Originally analog, it was an amazing feat for electrical engineers to find a way to transmit moving pictures over the air to viewers everywhere.

But people wanted more; bigger screens, increased clarity, zero static. Engineers squeezed everything they could out of the legacy TV format but analog TV just couldn't cut it, so they figured out how to digitize a video signal, and DTV was born. Suddenly in 2009 the world changed. The boob tube was discarded and everyone had flat panels with High Definition pictures! And not even that was enough so they soon doubled the HD to get 4K UHD. And now there are even 8K TVs (though there’s no content at that resolution.

But here you are with a box of old home videos, and if you can find a VCR to play them, they look really fuzzy on the big screen because they have, at most, ¼ of the resolution of the HD material that comes off air.  What can be done?

Enter AI at Advent Digitizing.

We now have the capability of “uprezzing” the SD footage to HD, so it looks almost as clear as broadcast HD.  The process of uprez handles Three important functions:

• Deinterlace: SD video is “interlaced.” To be able to transmit over the air efficiently, analog SD video divides the screen in 525 “lines,” which scan from top to bottom of the screen in 1/30^th of a second. However they split it in two, so that the ODD number lines appear in the first 1/60^th of a second, and the EVEN number lines appear in the second 60th of a second. This flicker is so fast the eye interprets it as smooth motion (unless the talent is wearing tightly patterned clothes, in which case they jitter). But digital video is “progressive,” meaning it draws all rows of pixels in order from top to bottom of the screen all at once. Depending on the TV's interpolation algorithm, you see jiggly artifacts on diagonal edges, or odd "comb" effects with rapid motion. To effectively uprez, the AI model smooths out the difference between pixels and eliminates the interlaced artifacts.
• Upsample: Now we play the numbers game. When digitized, SD video has 720 columns of pixels (picture elements) across and 480 rows tall (720x480). But HD digital video has 1920 pixels across by 1080 pixels tall. If we simply “scale up” the video as your big screen TV would do, it simply enlarges the pixels, leading to soft edges, stair step lines and an overall “fuzzy” look. The AI model searches for edges and faces in the footage and attempts to better define them, so the result looks much cleaner and sharper.
• Aspect Ratio Conversion: SD video is 4:3 “narrow” screen, but HD video is widescreen 16:9. We have two options. One is to scale the video so the vertical dimension fits the height of the HD frame and leave black “pillarbars” on the sides. The other is to crop top and bottom of the image and fit it to the width of the HD frame. For scenery with sky and grass, this usually works fine, but if the shot includes peoples’ heads very near the top of the SD frame, the heads may get cut off. We'll offer you the choce between pillarbox and crop-to-full.

The best thing about the Artificial Intellengce model we chose is that it is wholly local. It does NOT transmit your data anywhere at all.  It's all processed in our video editing machine, right here in Columbus, Ohio. Your images never touch the internet unless you request digital upload delivery.

The AI conversion does really amazing things.  Have a look at these samples from DV tape masters: (Use the Full Screen button, lower right, to see the subtle detail.)

We do something most other transfer companies don't: we transfer all video first to the i-frame DV codec as a "source file." The source files are huge, at 25megabits per second data rate, yielding files that are about 15 Gigabytes per hour! While those are too big for most media players, they're perfect for the uprez process, because each frame contains the entire picture, not a made up index of changes from the previous frames, as in MP4 compressed video. It always yields excellent results, provided the content is suitable.

There are a few things that may disqualify a tape for HD conversion:

• Noisy or slow-speed VHS footage.  If the tape is noisy with lots of "grain" from either filming in low light or using slow SLP mode, the AI won’t know what to do with all the noise, and you’ll see odd artifacts in the uprezzed version.
• Jittery fast camera movement. The AI is looking for edges to straighten. If it can find none, you’ll get oddball results.
• Highly detailed images, such as foilage. The AI may interpret this as noise and make it noisier.

Before making the HD uprez, we’ll examine the footage and maybe even test some footage to make sure it will be a successful transfer.

Uprezzing footage is a long process, even on a powerful computer, so the fees are high. But if the footage is precious and suitable, you’ll be able to enjoy the legacy footage on your HDTV for years to come.

Uprez your footage today at Advent Digitizing.

Get a quote and ask if HD Uprez is right for you!

---

---