- Abstract:
- Evolving object-based video coding standards, such as
MPEG-4, permit arbitrary-shaped objects to be encoded and decoded as separate
video object planes (VOPs). The MPEG-4 verification model (VM) proposes
that motion compensation (MC) based predictive coding is applied to each
VOP. As with all block matching MC schemes, the prediction quality is dependent
on each block representing an area of uniform translational motion. This
is rare for both frame-based real image sequences and arbitrary-shaped
objects. One solution is to allow the dimensions of blocks to adapt to
local activity within the object, larger blocks being used in large areas
of uniform motion, and smaller blocks where the movement is localised or
complex. In this paper we consider new developments for object-based systems.
Our techniques are as computationally efficient as conventional fixed size block matching and yet provide better quality prediction. This permits adaptive bit allocation between the representation of displacement and residual (error) data, and the variation of the overall bit allowance on a VOP-by-VOP basis.
The performance of our new object-based motion estimation algorithm is compared with conventional techniques and results are presented for complete MPEG-4 image test sequences, each comprising 300 frames. Significant improvements are shown in the number of motion vectors required to represent the motion components of a video object.
 |
 |
