Image Compression with a Vector Speck Algorithm

SPIHT is an efficient image compression algorithm based on zerotrees. The significant wavelet coefficients are located by a series of set partitioning operations and then scalar quantized. Block-based algorithms inspired by SPIHT such as AGP and SWEET have good performance, but they are not embedded. Pearlman et al. proposed a block-based SPECK algorithm using set partitioning of embedded blocks to exploit the energy clustering characteristics of the coefficients while the bit stream remains embedded. We here propose a variation on SPECK using vector quantization to code the significant coefficients. Different VQ techniques including TSVQ and ECVQ are also considered. Vector SPECK shows a performance improvement over JPEG 2000 at the cost of added complexity