Comparison of wavelet and cosine basis for representation of arbitrarily shaped image segments

Abstract

The efficient representation of arbitrarily shaped image segments is an important issue in object-based image coding. This paper describes an algorithm by which the given image segment is successively approximated using 2-D shape-independent basis functions defined on a rectangle circumscribing the image segment. Discrete cosine and discrete wavelet basis were used in the system respectively, and performance comparison between them was made. Simulation results show that wavelet basis usually yields better approximations than the discrete cosine basis. However, using wavelet basis, the extrapolation outside the image segment is not smooth, due to the selection of basis functions with small overlap with the image segment. To solve this problem, we proposed a new basis selection method which gives priority to basis functions that have larger overlap with the image segment. The extrapolation outside the segment is improved without degrading approximation quality.