University of Rochester
A Position Statement for Panel 2: Video representation,
1998 International Workshop on Very Low Bitrate Video Coding
Video representations for compression evolved from pixel-based to block-based
to object-based as digital video applications grew more sophisticated.
While block-based representations prove sufficient for the purposes of
compression efficiency, they are blind to content; and hence, do not allow
content-based interactivity. Object-based representations allowed by VRML
and MPEG-4 enable more sophisticated content-based interactivity, such
as tuning of compression parameters to each object, composition of content
from various sources, and object-based manipulation of content. Together
with temporal hierarchies, such as scenes and shots, we are now better
equipped for content-based video abstraction. However, the state
of the art in multimedia signal processing and computer vision precludes
the possibility of a high level abstraction fully automatically. In order
to advance the state of the art, we need to address the following questions:
are the presently known representations for visual objects sufficient for
content-based abstraction? How much can be done automatically? What kind
of interactive tools are needed to facilitate semi-automatic video abstraction?
Can effective video abstraction be coupled with compression methods to
improve compression efficiency?