Improvements to SPHIT for lossy image coding

Proposes five measures to improve the SPIHT (set partitioning in hierarchical trees) algorithm for lossy image coding. First, a new type of tree with virtual root is introduced to hold more wavelet coefficients. Second, an extra matrix is used to speed up the judgement of the significance of trees. Third, a pre-processing is done to smooth the coefficients before SPIHT encoding. Fourth, some predictable bits are omitted from the encoder output by rearranging the coding procedure. Finally, the quantisation is offset from the middle-point according to the statistics. Our experiments show that these improvements increase the PSNR (peak signal to noise ratio) by up to 5 dB at very low rates, and the average improvement at 0.2-1 bit/pixel is about 0.5 dB for the standard test images used. Despite the performance gain, the computational complexity is also reduced