Universal context modeling for lossy wavelet image compression

In this paper we present a new wavelet image coding scheme based on universal context modeling. The wavelet coefficients arc quantized using a layered quantization approach. The resulting symbol stream is entropy-coded with a context-based arithmetic coder. The efficiency of arithmetic coding depends on how well the probability model used by the encoder fits the data generating source. We apply the universal context mod-cling algorithm introduced by Rissanen and Weinberger to estimate both structure and parameter of the source model. This context selection strategy avoids the problem of context dilution which arises when using a plain Markov model and leads to higher compression.