On the Spiht-Based Multispectral Image Compression

In multispectral image compression, it has been recently shown that 3D-SPIHT (set partitioning in hierarchical trees) outperforms the benchmark JPEG-2000 multi-components. Despite the fact that the extended SPIHT algorithm to 3D provides the obvious way to process a multispectral image as a volumetric block and; consequently, maintains the attractive properties exhibited in 2D (good performance, low complexity and embeddedness), the adopted 3D trees structure still lacks in efficiency when applied to 3D wavelet transformed (DWT) multispectral images. Indeed, because of the texture nature of the spatial information content and the variety of frequency ranges in the consecutive bands, significant coefficients are more likely to be found in the high frequency wavelet subbands along both the spatial and spectral directions, especially when the value of the threshold decreases. As a consequence, the fact that each parent has 8 children in the 3D structure of SPIHT makes the list of insignificant sets (LIS) and the list of insignificant pixels (LIP) dramatically grow since each set is split producing 8 other sub-sets in case the parent has at least one non direct descendant significant with respect to the current threshold during the sorting pass. Thus, a large amount of bits is wastedly spent to sort the insignificant information. This paper discusses the use of 3D-SPIHT for multispectral images through an investigation based on results analysis. Then, we demonstrate that a straightforward 2D-SPIHT technique, suitably adjusted to maintain the scalability and carried out in the 3D-DWT domain, surprisingly yields a better performance