Game Development Reference
In-Depth Information
segmentations may be identified as some objects for further operation, e.g., coding
and tracking (Musmann et al.
1989
; Salembier et al.
1997
), which is called object-
based coding.
The fundamental difference between block and object-based coding schemes lies
in that the object-based coding structure concentrates on analyzing and synthesizing
the objects in an image (Salembier et al.
1997
), which brings several advantages over
block-oriented schemes, e.g., adaptation to the local image characteristics and object
motion compensation as opposed to blockwise motion compensation. Therefore, by
transmitting the object information, the conventional coding artifacts induced by
block-based coding scheme such as blocking and ringing can be avoided, and low
bit rate with high quality compression can be achieved. A number of remarkable
research activities on object-oriented image and video data compression have been
reported inHötter (
1990
), Hotter (
1994
), Gerken (
1994
), Ostermann (
1994
). A typical
object-based video coding aims at efficiently coding specific objects, e.g., a human
face. This kind of video sequence usually consists of stationary background scene
and dynamic foreground scene, such as telephony video, which is also called as
head and shoulder video sequence. Head and shoulder video compression has wide
applications in video conferencing and video phone, which requires low bit rate
video compression to represent the dynamic scene. To achieve higher efficiency
compression of audiovisual information with relatively low bit rate, MPEG started
to develop the international standard MPEG-4. In a word, object-based coding is one
of the most important milestones in the developments of video coding. It symbolizes
that the content aware coding technique comes to be a core technology in the model-
based coding framework and as important as statistical prediction/transform coding
methods, and it also inspires the following knowledge, semantic-based coding and
perceptual coding, etc.
In a sense, knowledge-based video coding is a kind of expansion of object-based
video coding, which encodes the object based on the knowledge of known objects,
e.g., face and body, etc. Generally speaking, the coder tries to recognize objects like
faces in a video sequence using the image/video analysis system, and then conducts
efficient coding by the corresponding prior knowledge about objects to be coded.
Take face coding as an example. First, the coder detects human face automatically
and identifies the locations of human face in video. As soon as the coder recognizes
the face, the coder switches the generic object-based mode to the knowledge-based
mode, i.e., face mode. Then, by incorporating face modeling into coding system,
the coders only need to encode the required analysis parameters of human face,
thereby saving bit rate significantly. The decoder can obtain human face by the
coded parameters using the same face modeling system. Therefore, the key is how
to model human face effectively. Many works are proposed to design algorithms for
face modeling efficiently and accurately (Aizawa et al.
1989
; Tekalp and Ostermann
2000
).
Different from the conventional video coding algorithms that encode each frame
of a video sequence using joint information of the current image signal and a pre-
dicted image signal, semantic video coding describes a video sequence using model
objects with behavior that represent real objects with their behavior (Wang et al.
