Lossy Hyperspectral Images Coding with Exogenous Quasi Optimal Transforms

Abstract

International audienceIt is well known in transform coding that the Karhunen-Loève Transform (KLT) can be suboptimal for non Gaussian sources. However in many applications using JPEG2000 Part 2 codecs, the KLT is generally considered as the optimal linear transform for reducing redundancies between components of hyperspectral images. In previous works, optimal spectral transforms (OST) compatible with the JPEG2000 Part 2 standard have been introduced, performing better than the KLT but with an heavier computational cost. In this paper, we show that the OST computed on a learning basis constituted of Hyperion hyperspectral images issued from one sensor performs very well, and even better than the KLT, on other images issued from the same sensor

Discrete Wavelet Transform and Optimal Spectral Transform Applied to Multicomponent Image Coding

“…Indeed, it is well-known that providing the mean square error as one distortion only is not sufficient to assess the quality of a codec for hyperspectral images (Christophe et al, 2005). However in the simulations presented in (Akam Bita et al, 2008;2010c;Barret et al, 2009) when the VM9 is used, the computational complexity of the EBCOT coder associated with its PCRD optimizer is very high, and when the BPE is applied to encode each component of the transformed image, the complexity of the algorithm for optimal allocation between components is also very high. In both cases, the computational complexity is too high for a compression system on-board a satellite.…”

“…The main drawback of the OSTs is their heavy computational cost, which is much higher than the one of a KLT or JADO (which both have roughly the same complexity). In order to reduce the complexity of a codec based on OrthOSTs, the authors of (Akam Bita et al, 2008;2010c;Barret et al, 2009) used the same strategy as in (Thiebaut et al, 2006): they replaced the OrthOST, which must be computed for each new encoded image, with an exogenous quasi optimal spectral transform. This last transform is an OrthOST computed once and for all on a learning basis constituted of images from only one spectrometer and which is then applied to any image to be coded stemming from the same spectrometer.…”

Lossy and lossless compression of MERIS hyperspectral images with exogenous quasi-optimal spectral transforms

“…The main draw-back of OST and OrthOST is their computational complexity, which is much higher than the KLT one [10], which is roughly the same as JADO. Thus, in [17][18][19], the authors of that paper adopted the above strategy leading to the exogenous KLT, replacing KLT by OrthOST, in order to compute a fixed quasi optimal spectral transform on a set (the learning basis) of images coming from one spectrometer (and only one). They showed, on images extracted from Hyperion hyperspectral images retaining only 45 bands (visible and near infrared) of a single imaging spectrometer, that good performance can be achieved in lossy compression with the exogenous OrthOST, using either the JP2K codec called Verification Model [20] version 9 (VM9) or the Bit Plane Encoder [21,22] (BPE) of the CCSDS.…”

Regression Wavelet Analysis for Lossless Coding of Remote-Sensing Data

“…Yet a third approach consists in learning the transform on a set of images of one particular sensor in order to obtain an efficient transform that can be applied to new images from the same sensor [18]- [21].…”