|
|
Posted by Jukka Aho on 11/15/06 07:42
Richard Crowley wrote:
>>>> Interesting that say converting from NTSC to PAL is best? I've
>>>> always found it better the other way. It's always better to down
>>>> convert from a superior format than try to add data upwards.
>>> Spatial interpolation is significantly easier than temporal.
>> That's true as a general statement, but does it work as an answer to
>> the question? I'd argue that for a perfect PAL <=> NTSC conversion
>> you will have to do temporal interpolation regardless of the
>> direction of the conversion.
> There is no "perfect conversion".
Some conversion methods are more perfect than the others. :)
> However going from 30FPS to 25FPS at least gives you more to work with
> than you need. Going in the other direction requires the
> hardware/software to "use its imagination" to fill in the missing
> pieces. Which one do you think is easier to do?
It's easy to fill in the missing pieces by duplicating frames. Dealing
with the excess by doing the converse (dropping frames) is equally easy.
Both these simple methods lead to a similar kind of distortion: aliasing
in the temporal dimension. Motion that was steady and smooth in the
original clip no longer isn't: it has a beating discontinuity at those
points where frames were duplicated or dropped.
The only way around this problem, except for slowing down or speeding
up, is isolating the moving objects/scenes, calculating motion vectors
for them, and synthesizing new frames out of thin air which depict
moving objects at locations where they must have been in-between the
original frames, but which were never actually shot with a camera.
--
znark
Navigation:
[Reply to this message]
|