red darf Pal>NTSC how did they do it?

[duhmez]

I bought the first 4 seasons of red dwarf on dvd which natively ere pal and interlaced. My versions are NTSC. I analyze the frames and I see 29.970 distinct frames per second. they are not using puldown. Each frame looks very crisp and clean (interlacing artifacts present of course though) the videos look great but I was wondering how the hell they did the conversion and have it look so good, and be 29.970 individual frames per second. when homeboy is moving his arm, there is no doubling of frames, no noticeable blending or anything.

[miamicanes]

I haven't seen it (actually, I haven't seen Red Dwarf itself, either), but my guess is that the producers *might* have hooked up with one of the few companies involved with the development of "motion vector temporal rate conversion" (MVTRC).

It's something that Microsoft has been quietly backing for some time now (do a search for "temporal rate correction" at MSDN). Basically, it takes advantage of the fact that the cel animation metaphor (animated, moving "sprites" over a panning "background") can be applied to real video too if you've got enough processing power.

Grossly oversimplifying the theory a bit:

* deinterlace the image.

* do edge-detection to find objects that seem to be moving relative to their surroundings.

* classify the remaining area as "background". Analyze a few frames before and after to determine its acceleration and velocity for the current frame.

* Move the background to the location predicted by its acceleration and velocity, then study the adjacent frames (or several) to "fill in the gaps" that were obscured by the edge-detected moving objects in step 1.

* Do the same thing with the edge-detected objects. Determine their acceleration and velocity, and move them to the location predicted by it.

* Now study the edge-detected objects a little harder to figure out how they might have morphed between the current frame and the next, and predict their appearance at that point in time between the two.


Now, in the steps above, I hinted that we're decomposing and analyzing the image to predict its appearance halfway between the current and next frame. Actually, you can use the technique to interpolate any arbitrary number of frames (within reason). Going from 540i50 Pal to 480p60 ATSC, you'd grab 5 deinterlaced frames at a time. Pal Frame #1 would correspond to 480p60 frame #1... but 480p60 frame #2 would have its velocity and acceleration-predicted appearance roughly 20% influenced by Pal frame #1, and 80% influenced by Pal frame #2.

This is hardcore, high-end cutting-edge stuff. But hopefully, it's going to become increasingly common.

[oddbod77]

"Motion Vector Steered temporal rate conversion" cripes now that's a mouthfull. High end plasma TV's already do this which is one of the reasons they are so expensive.

A far simpler (and cheaper) way would be the tried and true "speed it up and resample the audio" method. You can do this at home by following this guide.