A compressive sampling scheme for iterative hyperspectral image reconstruction

Andrea Abrardo, Mauro Barni, Cesare Maria Carretti, Simeon Kuiteing Kamdem, Enrico Magli

Abstract: Compressed Sensing (CS) allows to represent sparse signals through a small number of linear projections. Hence, CS can be thought of as a natural candidate for acquisition of hyperspectral images, as the amount of data acquired by conventional sensors creates significant handling problems on satellites or aircrafts. In this paper we develop an algorithm for CS reconstruction of hyperspectral images. The proposed algorithm employs iterative local image reconstruction based on a hybrid transform/prediction correlation model, coupled with a proper initialization strategy. Experimental results on raw AVIRIS and AIRS images show that the proposed technique yields a very large reduction of mean-squared error with respect to conventional reconstruction methods.