-
Subpixel Image Tracking for Image Stabilization: How do I avoid “subpixel softening”?
Hello,
I was reading reading the Encyclopedia of Visual Effects (Allen & Connor), and came across the section in “Sub-pixel Image Tracking.” It says, “…subpixel resolution should be set to 1/256 or greater…”
A “Concatenation Workflow” is offered, in which clip is stabilized, comps and matchmoves are added, then the clip is destabilized back to it’s original shakiness. But, I could not find a way to avoid subpixel softening when stabilizing footage and keeping it stabilized.
However, I do not see a way (and cannot conceive of one) to only stabilize a shaky shot (and keep it stabilized), without inducing considerable resolution degradation to the image. I’m thinking that, if one pixel has one discrete, independent piece of information (I know this is not usually the case, but say it’s RED 4K footage, downsampled into 1920 of horizontal lines), then image stabilization would seem to decrease resolution by a factor of four. Along this line of thinking, this would be because stabilizing any horizontal movement between two frames requires that one pixel is combined with at least one neighboring horizontal pixel, and stabilizing any vertical movement between two frames requires that each pixel is combined with at least one other vertical pixel. If shot shakes horizontally and vertically between two frames, it would seem that one pixel would be combined with the neighboring 3 pixels.
1) What’s the work around?
I there one?
Maybe track, crop, then uprees slightly with something like Red Giant Instant HD (I think this plug-in works works a little like PS plugin Genuine Fractals/Perfect Resize 7).
Surely this is not the workflow you pros use.
2) Is there an AE plugin to do something like Genuine Fractals/Perfect Resize 7 does, without necessarily resizing (find “lines” and draw them in a way that’s different from sharpening)?
3) For those of you who are are also Apple Motion artists, does Motion have some sort of magic way of handling this, that AE does not?
I will be checking this thread, but will not have time to give a proper reply until Friday.
Thanks in advance for your time.
Sincerely,
Chris -
1 Reply
-
[David LaRonde]
“YOU would know that. But After Effects, not being a clairvoyant application, would not. Before you tracked the motion, it would see footage at 1920×1080 pixels. It would not know a blessed thing about its previous life as 4K footage, even if you tried to tell it…”
Oops. I just meant to make the question clear by using “perfect” footage as an example — 1920×1080 4444, with each pixel being able to resolve a single independent discrete piece of information.
No such camera outputs footage of this clarity, to my knowledge, so the nearest example I could think of was 1920×1080 footage, downsampled from the 4K RED. But I did not mean to suggest I would be using a proxy workflow, so the idea of 4K just confuses the question.
Hope that clears that up. Seems like the more I say, the more vague my questions become… 😉
Reply to this Discussion! Login or Sign Up