Three-Dimensional Wavelet-Based Compression of Hyperspectral Images

Abstract

Hyperspectral images may be treated as a three-dimensional data set for the purposes of compression. Here we present some compression techniques based on a three-dimensional wavelet transform that produce compressed bit streams with many useful properties. These properties are progressive quality encoding and decoding, progressive lossyto-lossless encoding, and progressive resolution decoding. We feature an embedded, block-based, image coding algorithm of low complexity, called SPECK (Set Partitioning Embedded bloCK), that has been proposed originally for single images and is modified and extended to three dimensions. The resultant algorithm, Three-Dimensional Set Partitioning Embedded bloCK (3D-SPECK), efficiently encodes 3D volumetric image data by exploiting the dependencies in all dimensions. We describe the use of this coding algorithm in two implementations, first in a purely quality or rate scalable mode and secondly in a resolution scalable mode. We utilize both integer and floating point wavelet transforms, whereby the former one enables lossy and lossless decompression from the same bit stream, and the latter one achieves better performance in lossy compression. The structure of hyperspectral images reveals spectral responses that would seem ideal candidates for compression by 3D-SPECK. We demonstrate that 3D-SPECK, a wavelet domain algorithm, like other time domain algorithms, can preserve spectral profiles ¡

Randomized SVD Methods in Hyperspectral Imaging

“…For practical purposes, an error rate in the order of 0.001 might be sufficient, and this would result in a compression ratio of 2.5 to 4. For comparison purpose, the 3D-SPECK [7] on a small dataset of size 320 × 360 × 58 results in a compression ratio of 1.12 at the 16-bit coding. If more sophisticated coding algorithms than Hoffman coding are applied here, we could see more improvements on the compression ratios.…”

“…For lossless compression of HSI data, there have been efforts to exploit the correlation structure within HSI data plus coding the residuals after stripping off the correlated parts; see, for example, [7,8]. However, given the large number of pixels, these correlations are often restricted to the spatially or spectrally local areas, while the dimension reduction techniques essentially explore the global correlation structure.…”

Regression Wavelet Analysis for Progressive-Lossy-to-Lossless Coding of Remote-Sensing Data

“…Some schemes apply 3D transforms [1,2], commonly based on wavelets, to handle the joint correlation in both spatial and spectral dimensions. Other somewhat different schemes apply separately a 1D spectral transform followed by a 2D spatial transform, using the most convenient one in each dimension.…”

Effective compression of hyperspectral imagery using an improved 3D DCT approach for land-cover analysis in remote-sensing applications

“…However, Tang and Pearlman have proved that 3D SPIHT yielded higher SNRs, e.g. 1.5 to 3.5 dB, than JPEG2000 multi-component at various compression bit rates (Tang and Pearlman 2006b). Nevertheless, 3D SPIHT provided comparable results to JPEG2000 multicomponent as shown by Christophe et al (Christophe, Mailhes, and Duhamel 2008).…”