Motta, Rizzo, and Storer, the editors of this volume, are veterans in the field of data compression, both individually and collaboratively. They bring together a concentrated set of contributed papers, focusing on compressing hyperspectral (multidimensional) data. The growing interest in such complex datasets has been fueled by recent applications in remote sensing from air and space, a technology that has matured from a range of wavelengths combined into a single signal stream to hyperspectral (hundreds of bands) and ultraspectral (thousands of bands). Because of the size of the data, compression technology plays an important role. In addition to the obvious defense and security applications, this advance in imaging and sensory technology has been coupled with the availability of the data for academic and commercial platforms.

The first chapter provides a review of compression concepts. It also provides a suitable architecture for the compression of hyperspectral data. Data compression techniques are categorized depending on whether or not the original data can be exactly reconstructed from the compressed data. Lossless compression (chapters 2 to 6) provides for the perfect reconstruction of original data, but does not have the higher compression ratios of lossy encodings (chapters 7 to 12), where some information is discarded. Since the effects of this data loss cannot be completely evaluated a priori, lossless compression is still favored when storing and transmitting hyperspectral image data. Chapter 13 investigates the image artifacts that can arise from lossy compression.

Chapter 2 uses reversible integer wavelet transforms for compression. This approach is based on the fact that when an image is decomposed based on its frequency, the coefficients at a higher frequency are closely related to the coefficients at a lower frequency; then, the error residuals are encoded using a wavelet coder. Chapter 3 concentrates on band ordering and modeling. It presents a fast heuristic for the band ordering that utilizes a correlation factor to examine interband similarity, followed by three possible types of linear prediction models: adaptive, delta-coding, and clustered. Chapter 4 explores the differences between compressing hyperspectral image data and ultraspectral sound data based on the user constraints for a given application area. In chapter 5, the authors incorporate partitioned vector quantization, where each dimension of the data hypercube is treated as a single vector, into a compression algorithm. They use adaptive partitioning with a novel locally optimal algorithm to reduce the size of the source code alphabet. Chapter 6 applies (near) lossless compression to NASA space-borne imaging spectrometers for disseminating to users the underlying thematic information.

Chapter 7 focuses on three strategies for hyperspectral image classification: the first with a physical meaning, the second based on spectral unmixing, and the third as a hard classifier. Chapter 8 measures the impact of different band orderings on the compression ratios. The authors conclude that lossless compression ratios based on compressed file size outperform those based on predictive abilities, but not by much. The algorithms of chapter 9 employ a variety of decorrelative techniques along with various embodiments of trellis-coded quantization. Various coder configurations are considered, with each coder possessing a unique set of attributes that may be suited to a particular application. Chapter 10 presents compression techniques based on three-dimensional (3D) wavelet transforms that produce compressed bit streams with useful progressive properties: quality encoding and decoding, lossy-to-lossless encoding, and resolution decoding. Their algorithm is a 3D analog of the SPECK algorithm, called 3D-SPECK. Chapter 11 develops spectral and spatial compression techniques utilizing principal component analysis by modeling the virtual dimensionality for hyperspectral image compression to estimate the actual number of principal components necessary. Their proposed method provides a much better estimate than the commonly used criterion determined by the sum of the largest eigenvalues. Chapter 12 presents a nice application of compression of earth science data using the international standard JPEG2000.

This compendium describes cutting-edge compression technology, and is sure to occupy an important position in the current literature of the field. The editors have accomplished their goal of making this technology available to the educational and industrial communities.