Kevin J. O'Connell

Object-adaptive vertex-based shape coding method

The paper presents a new technique for compactly representing the shape of a visual object within a scene. This method encodes the vertices of a polygonal approximation of the objects shape by adapting the representation to the dynamic range of the relative locations of the objects vertices and by exploiting an octant-based representation of each individual vertex. The object-level adaptation to the relative-location dynamic range provides the flexibility needed to efficiently encode objects of different sizes and with different allowed approximation distortion. At the vertex-level, the octant-based representation allows coding gains for vertices closely spaced relative to the object-level dynamic range. This vertex coding method may be used with techniques which code the polygonal approximation error for further gains in coding efficiency. Results are shown which demonstrate the effectiveness of the vertex encoding method. The rate-distortion comparisons presented show that the techniques adaptive nature allows it to operate efficiently over a wide range of rates and distortions and across a variety of input material, whereas other methods are efficient over more limited conditions.