The Telecining Process
Start with 24 fps video (frame-based or “film”)
— Slow to 23.976 fps
— Break into fields
— Create 10 fields for every 8 (by duplicating and shuffling around)
Most movies are still shot at 24fps on actual film. For TV, they are converted to 29.97 fps because the American television system runs at that framerate (though more acurate is to say that it runs at a field rate of twice that, at 59.94 fields per second). This conversion from 24 to 29.97 is called telecining.
Some DVD’s store the original 24 frames/sec, whereas others store 29.97 frames/sec. When you run dgindex on your dvd, it scans the dvd and will determine if you have “FILM” or “NTSC”. If it’s film, you’re in luck and can move on to an encode. If it is NTSC (telecined), then the telecining process must first be undone using the aptly-named inverse telecine operation (or “IVTC” as it’s often called).
Why do we have to do this? Why not just encode it at 29.97 fps and save ourselves the trouble? It is possible to skip the ivtc, but we would get a major quality loss in doing so. First, telecining creates noticeable artifacts (the “combing” artifacts similar to true-interlaced video). Second, by its very nature of creating extra frames, it adds a “stutter” to the video making what would otherwise be smooth motion look jerky. Furthermore, from the encoding side of things, combing artifacts are harder to encode because they are sharp edges and the more sharp edges, the more bits required. And finally, after converting from 29.97 to 23.976 fps, there are 20% less frames to encode, which means all the remaining frames get that many more bits. So you see, it really becomes quite clear that Telecining is evil and must be destroyed at all costs.
How exactly does telecining do what it does?
The goal of telecining is to change video from 24 frames per second to 29.97 frames per second, which is another way of saying we need to somehow create 29.97 frames for every 24. The first step is to slow down the film speed from 24 to 23.976 (such a minor slowing is not noticeable by humans anyway). 23.976 is exactly 4/5th’s of 29.97 and so it now remains to create 5 frames where there once were 4.
Next, we break each frame into fields. 4 frames becomes 8 fields, and now the goal is to create 10 fields where there once were 8.
The solution is to simply copy 2 of the 8 fields and use them again to make a total of 10. For our example, we start with four frames:
A B C D fields
1 2 3 4 frame number
Then we break them into fields:
A1/A2 B1/B2 C1/C2 D1/D2 fields
1 2 3 4 frame number
Then we take our fields and create the following 5 frames using the pattern:
A1/A2 B1/B2 B1/C2 C1/D2 D1/D2 fields
1 2 3 4 5 frame number
We’ve now pulled a 5-frame rabit out of a 4-frame hat by duplicating both the B1 field and the D2 field. The price we pay is that 2 of our 5 new frames are now interlaced–they’re constructed from two different frames of our original film, every other line borrowing from each! Interlacing looks bad and is harder for an encoder to encode (it requires more bits, as mentioned above).
Note also that there is a sort of “stutter” introduced, though we often don’t notice it since the fields go by so quickly. Where each frame used to get 1/24th second, now frames B and D (what’s left of them, anyway) are 50% longer, though they’re mixed with the other frames. And as for frame C, it is flat out gone–half of it is in the 3rd of our new 5-frame collection, and half of it is in the 4th.
Now that we know how telecining works, it appears the process should be reversible and thankfully, it is. TMPGenc has a nice applet for automatic ivtc, and avisynth has several intelligent, automatic ivtc plugins available for it. Both also have manual implementation of IVTC (which is easier than it sounds, but the automatic methods are not bad at all).