Dimensionally Reduction and Recovery for Hyperspectral Images using Discrete Prolate Spheroidal Sequences

Seda Senay and Roummel F. Marcia


Hyperspectral images, principal component analysis, discrete prolate spheroidal sequences, dimensionality reduction


In hyperspectral imaging systems, principal component analysis (PCA) is the conventional way of spectral dimensionality reduction. Also known as the Karhunen Loeve Transform, PCA is a data dependent transform which is obtained from the eigendecomposition of the covariance matrix of the dataset of interest. The computational burden of the PCA transform often exceeds the capacity of resource constrained hyperspectral sensing platform. We propose to use discrete prolate spheroidal sequences (DPSSs) for the dimensionality reduction of spectral bands. DPSSs construct a highly efficient basis that captures most of the signal energy. Recently, a compressive projection PCA (CPPCA) method has been proposed for dimensionality reduction of spectral bands and recovery of the image without having the computational burden of conventional PCA transform. We show that DPSSs can be used efficiently for dimensionality reduction and recovery of the hyperspectral image data and its performance exceeds the CPPCA.

Important Links:

Go Back