TEXTURE CODING ON ARBITRARILY-SHAPED IMAGE SEGMENTS - TRANSFORM METHODS

The new generation of image sequence coders - rile object-oriented cod ers-consist in describing the sequence as ii set of moving objects, of arbitrary size and shape. This approach allows rite development of ne w object-based functionalities. One of the main problems to be solved in region-based coding is the coding of the texture of the objects. As for the classical block-based coding, a linear transform can be used to eliminate the input signal redundancies. This article describes tra nsform methods for arbitrarily shaped regions of support, classified i nto two categories : the shape-adaptive methods and the extrapolation methods. The main differences between these two classes are discussed and the points that need further investigation are raised.