Efficient Address Generation Sub System for SPIHT based Memory- Efficient Image Compression Scheme

We present a hardware model of efficient address generation sub system for SPIHT in our unified model of image compression optimized for memory and throughput. The unified model utilizes Spatial Prediction (SP) and Set Partitioning In Hierarchical Trees (SPIHT). A single re-usable list is used in the proposed modified SPIHT algorithm, instead of three continuously growing linked lists existing in basic SPIHT algorithm. Augmented with exchange of significant and refinement passes, this leads to memory optimization at the algorithmic level while maintaining Peak Signal to Noise Ratio (PSNR) performance at lower bit-rates. The modified SPIHT algorithm yields reduction of memory requirements of the order of 65.16% on an average as compared to standard SPIHT. Effective 1D scheme for coefficient addressing is used in SPIHT modeling. The speed of accessing the coefficients is increased by 66.67% and throughput is increased by 61.11% using 1D addressing scheme as compared to 2D addressing scheme.