Maximum Distance-based PCA Approximation for Hyperspectral Image Analysis and Visualization

Principal Component Analysis (PCA) is a widely-used statistical tool for multivariate signal analysis, including optimized representation, compression and visualization in the case of hyperspectral satellite images. We propose a geometrical construction of an approximation of the PCA based on the maximum distance instead of the classical statistical approach based on computing the eigenvectors of the data covariance matrix. We validate our approach on synthetic 2D data with Gaussian distribution and prove its usefulness for visualization of hyperspectral satellite images. We discuss our results and draw the conclusions.