Representation of arbitrarily-shaped image segments using wavelet basis

In object based image coding, an important issue is the efficient representation of arbitrarily shaped image segments. The paper describes a representation method using selected discrete wavelet basis. In this algorithm, the given image segment is successively approximated using 2D shape-independent discrete wavelet basis functions defined on a rectangle circumscribing the image segment. Simulation results show that this algorithm gives a better approximation result than an algorithm using a discrete cosine basis. However, extrapolation outside the contour is not smooth due to the selection of small energy basis. To rectify this problem, we proposed a new basis selection method which gives priority to the basis functions that have a basis only in a smaller region defined by a coefficient selection mask (CSMask), thus the extrapolation result is obviously improved