Wavelet-Based Image Segment Representation

Introduction: An important issue in object-based image coding is the efficient description of image segments having arbitrary shapes [1]. Kaup described an algorithm based on Matching Pursuits (MP) [2] to represent arbitrarily shaped image segments by selecting a small set of discrete cosine basis from a larger basis set defined on a rectangle circumscribing the region[3] (we denote it as MP-DCT). Due to the good energy compaction property of the discrete wavelet transform (DWT), we used this basis in the MP framework to formulate the MP-DWT. And we found an efficient basis selection method by restricting the basis functions in an area defined by Coefficient Selection Mask (CSMask). Simulation results showed that our algorithm gives good approximation of the given segment. To further reduce the computational load, smaller basis selection ranges are defined.