Adaptive two-stage Karhunen-Loeve-transform scheme for spectral decorrelation in hyperspectral bandwidth compression

摘要

A computationally efficient adaptive two-stage Karhunen-nLoeve transform u0001KLTu0002 scheme for spectral decorrelation in hyperspectralnlossy bandwidth compression is presented. The component decorrelationnof the JPEG 2000 u0001extension 2u0002 is replaced with an adaptive twostagenKLT scheme. The data are partitioned into small subsets. Thenspectral correlation within each partition is removed via a first-stage KLT.nThe interpartition spectral correlation is removed using a second-stagenKLT applied to the resulting top few sets of equilevel principal componentnu0001PCu0002 images. Since only a fraction of each equilevel first-stage PC imagesnare used in the second stage, the KLT transformation matrices willnhave smaller sizes, leading to further improvement in computationalncomplexity and coding efficiency. The computation of the proposed approachnis parametrically quantified. It is shown that reconstructed imagenquality, as measured via statistical and/or machine-based exploitationnmeasures, is improved by using a smaller partition size in the first-stagenKLT. A criterion based on the components of the eigenvectors of thencross-covariance matrix is established to select first-stage PC images,nwhich are used in the second-stage KLT. The proposed scheme alsonreduces the overhead bits required to transmit the covariance informationnto the receiver in conjunction with the coding bitstream.