Page 65 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

42 Chapter 2. Video Coding: Fundamentals

coding o6ers a potential saving in bit rate, which makes it attractive for very-
low-bit-rate applications.
Model-based coding methods can be broadly classi ed as object-based or
knowledge-based. Object-based coding methods deal with unknown (arbitrary)
objects, whereas knowledge-based coding methods assume a priori knowledge
of the objects being modeled (e.g., a 3-D wireframe face model is usually
employed for head-and-shoulders sequences typical of videophone applica-
tions). Knowledge-based coding methods are generally successful in tracking
the global motion of the object (e.g., rotation and translation of the head),
but su6er from errors in estimating local motion (e.g., the movement of the
eyes, lips, and so on). Semantic-based coding is a subset of knowledge-based
coding methods that models local motion using a set of action units (e.g., a
combination of facial action units can lead to a given facial expression).
Despite their good performance at very low bit rates, model-based coding
methods have their problems. For example, at lower bit rates, the analysis and
modeling processes become more complex and the model needs to be more
object speci c. In addition, the analysis and tracking methods usually require
some degree of human intervention or some a priori assumptions about the
nature of tracked objects. Another problem is that, in some cases, severe or
sustained failure of tracking or modeling may occur, leading to an increase in
the bit rate or a deterioration in the video quality. However, continuous re-
search e6orts in this area are addressing such problems. For example, switched
model-based coders, with a fallback mode to conventional coding, have been
proposed to solve the problem of model or tracking failure [56]. For a good
review of model-based coding, the reader is referred to Ref. 57.