Re: Zero motion vector Inter prediction in H264
- From: "George Johnson" <matrix29@xxxxxxxxxxx>
- Date: Fri, 22 Aug 2008 08:52:54 -0400
<dipumisc@xxxxxxxxxxx> wrote in message
news:5d8e491b-0450-4b53-a2a7-cbf32d5b1bd2@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
I want to do an H264 encoder with minimal inter prediction. The first
choice is doing Intra frames only. I can find some reference to
quality vs bitrate for H264 Intra-only profile on the net. The next
step I am thinking is do a differential inter prediction. Basically
assume a (0,0) motion vector for every macroblock. Is that going to
result in reduction of bitrate from Intra-only encoding? I am looking
for research results that shows typical gain of such an inter
prediction over standard test sequences.
Any other suggestion on improving compression efficiency from Intra-
only with minimal additional processing requirements? Of course the
assumption is that the quality is the same.
Thanks.
-Dipu
Easy suggestion. Allow "Layer Clumps".
In a movie where filming truck is driving parallel with a car is driving
down the other side of the road, the background moves, but the car itself
just tends to jiggle back and forth in frame. Compressing the background as
a subordinate layer to the car layer would give you better compression
results than trying to compress the whole series of frames "as is".
Allowing motion vector layers to slide over each other will give you
superior results in general than just going with temporally-tracked vector
spline clumps. The simplest way to keep the memory costs down would be to
create a vector outline of your clumped object layer and then fill that in
with your temporally-active bitmaps. This is, of course, logically inferior
when used on complex objects spinning on axis in 3D and in regards to mostly
transparent layers and generally action-engaged human figures.
Better results can be gained with decent math processing chips and a bit
more memory and computational time.
Cheap consumer electronics tend not to have lots of hardware memory or
lots of computational time to spare though.
.
- Prev by Date: Re: XDepth® High Dynamic Range Video Compression Technology on the Market
- Next by Date: How to tell when your great compression idea has gone bad
- Previous by thread: Does PKZIP use zlib?
- Next by thread: Re: Zero motion vector Inter prediction in H264
- Index(es):
Relevant Pages
|
