High-quality image compression using a wavelet transform and left-neighbour-parent dependencies
F. Lesani, George H. Freeman · 2002
Image coding is performed using a wavelet transform followed by coefficient quantization and entropy coding. Coefficients which are below a band-dependent threshold are set to zero, and those remaining are uniformly quantized with a band-dependent number of steps. Each quantized coefficient (including zeros) is placed in a context class depending on its left neighbour in the same band and its parent in the corresponding band at the next coarser level of resolution. The coefficients in each context, and the low-low-band image from the coarsest level of resolution, are separately entropy coded by an adaptive arithmetic code. Results for specific images show the effects of the quantizer design and demonstrate a reasonable extension of existing wavelet-transform-coder performance to a region of higher quality (about 35-50 dB of peak SNR).