Efficient coding of sparse trees using an enhanced-embedded zerotree wavelet algorithm

Abstract

In the Embedded Zerotree Wavelet (EZW) algorithm, a large number of bits are consumed in the encoding of Isolated Zero (IZ) symbols. This is the main bottleneck of the EZW algorithm, which limits its performance in terms of compression gain. To circumvent this limitation of the EZW algorithm, we propose in this paper, the Enhanced-EZW (E-EZW) algorithm based on the novel concept of a sparse tree (ST) encoding scheme. The ST encoding scheme provides an efficient encoding of ‘IZ’ symbols and eventually gives significant improvement in compression gain. Image features are clustered at various locations in an image, which gives rise to spatial correlation between Significant Coefficients (SCs) at these locations. Based on the above observation, we further propose differential coding of relative position of SCs in ST (DCORPS) in the E-EZW (DCORPS E-EZW) algorithm. We analyze cases where the ST coding gives higher coding gain compared to the EZW algorithm. Further, we see that DCORPS in sparse tree coding improves the overall coding efficiency of the E-EZW algorithm. By simulation results, we also demonstrate that the E-EZW and DCORPS E-EZW algorithms outperform two other important wavelet-based compression algorithms: namely set partitioning in hierarchical trees (SPIHT) and JPEG-2000 for a representative set of real-life images.