Adaptive hyperspectral image compression using the KLT and integer KLT algorithms

Abstract

The use of the Karhunen-Loéve Transform (KLT) for spectral decorrelation in compression of hyperspectral satellite images results in improved performance. However, the KLT algorithm consists of sequential processes, which are computationally intensive, such as the covariance matrix computation, eigenvector evaluation and matrix multiplications. These processes slow down the overall computation of the KLT transform significantly. The traditional KLT can only offer lossy compression; therefore, a reversible KLT algorithm, named the Integer KLT, is used for lossless compression. The Integer KLT includes more computational processes and, hence, it requires a longer processing time. The acceleration of these processes within the context of limited power and hardware budgets is the main objective of this paper. The computations of each of these processes are investigated thoroughly. Subsequently, a novel adaptive architecture for the computation of the KLT and the Integer KLT is proposed. The proposed system improves the traditional KLT performance compared with a previous architecture, and offers significant improvement for hyperspectral data with a larger spectral dimension. The experiments showed an overall improvement of up to 4.9%, 11.8% and 18.4% for 8, 16 and 32 spectral bands, respectively. In addition, this paper addresses novel hardware aspects of the Integer KLT implementation. The scalability of this hardware architecture can offer much higher level of parallel computing than processor platforms.