This book presents the investigative work on an advanced video coder
based on the structure of a classical predictive coding scheme,
substituting two basic components: a) motion prediction is made with
triangle meshes instead of blocks; b) transform coding is realized with
discrete wavelet transform instead of DCT. These parts have been
investigated separately by many scientists but never before in
combination. Within this framework, the author investigated the
individual processing of arbitrarily shaped regions with the aim of
preferential coding. Here, different types of regions were considered:
pre-segmented objects, and regions created with help of the triangle
mesh. The mesh regions extract skin-colored areas, or those with a high
prediction error to code them with preference in order to achieve
locally higher subjective qualities. Results show, that the combination
of mesh-based motion prediction and wavelet transform yields better
subjective and objective image coding quality than either of them in
combination with a block-based technique.