Reconstructing Hyperspectral Images from Compressive Sensors via Exploiting Multiple Priors

The compressive sensing technique is a new mechanism for hyperspectral imaging, and practical hyperspectral compressive sensors have been designed. To improve the reconstruction performance, compressive sensing should exploit some prior knowledge of the original signals. In this letter, we propose a novel and efficient reconstruction approach for compressive sensors by exploiting four important priors: spatial 2D piecewise smoothness, adjacent spectrum correlation, low rank, and structure similarity property. It is worth mentioning that the structure similarity property has never been taken into account in the existing schemes. In addition, an efficient solve algorithm is developed for our approach. The simulation experimentations show that our approach recovers hyperspectral images with far less reconstruction error than existing schemes.