Reduced Bit-Rate Uniform Quantization for SPIHT Encoding X-W. Yin, M. Fleury, and A. C. Downton Department of Electronic Systems Engineering University of Essex, Wivenhoe Park, Colchester, CO4 3SQ, U.K. tel: +44 - 1206 - 872817 fax: +44 - 1206 - 872900 e-mail fleum@essex.ac.uk Abstract SPIHT is a scalar quantizer for images that have undergone subband decomposition by wavelet transform. In this letter, reduced bit-rates have been achieved by merging the first and second passes of the already efficient SPIHT algorithm. The modification requires neither entropy encoding nor significant increase to the algorithmic complexity. 1 Introduction Set Partitioning in Hierarchical Trees (SPIHT) [1] is an established method of uniformly quantizing the coefficients of a wavelet-transformed image. SPIHT progressively transmits bit-planes in successive passes, included an embedded bit-map of significant coefficient sets with each pass. SPIHT compression is achieved [2] by coding sets that share a common characteristic with a single bit in the bit-map. The choice of aggregate set is made via quad- tree structures according to the structure of input images and the transform itself. This letter shows that for many natural images a modest bit-rate improvement can be achieved by including a per-coefficient bit when there are few coefficients that are significant (likely to be set) in the first pass. 1