Wavelet coding using pruned trees and vector quantization

We present here a novel coding approach in the wavelet transform domain, based on a pruned tree structure of the transformed signal. The approach is based on the observation that on one hand a large portion of the wavelet transform samples may be nullified without significant loss of reconstruction quality, and on the other hand samples with relatively high energy tend to concentrate around local transients of the source signal. A pruned tree is built in which each node consists of a cluster of samples of the same scale which are centered at a local transient. The branches of the tree connect nodes of adjacent scales. The tree structure is coded by a simple binary code. A vector quantizer is used to code each node. Compression ratios of 1:25 on a gray-level images are achieved with good quality