Multiresolution spectral imaging by combining different sampling strategies in a compressive imager, MR-CASSI

Abstract

The Coded Aperture Snapshot Spectral Imaging system (CASSI) is a remarkable architecture based on CS theory which senses the spectral scene by using two-dimensional coded focal plane array (FPA) projections. The CASSI system can be characterized by a measurement matrix which can be designed according to the requirements of a particular issue. Traditional approaches require recovering all the data with high resolution which involves a large amount of data and in consequence high costs of transmission and storage. However, in several applications, the data analysis is focused on only specific regions of the images. Therefore, this work proposes a multiresolution compressive architecture (MR-CASSI). MR-CASSI is focused on the most important spatial or spectral areas of the scene to be analyzed without background subtraction, allowing to reduce the amount of data preserving all the scene, selection of these areas of interest is pre-selected. The MR-CASSI is designed from a measurement matrix, such that the system samples the scene to recover multiresolution images low resolution for the background and high resolution for the spatial target or spectral regions. An important aspect of this proposal is that we can estimate multiresolution images without extra processing. From simulation results for the MR-CASSI architecture, it was found that compared to a traditional system, our approach overcomes an average 12dB of PSNR with a low-resolution system by using different decimation factors to obtain multiresolution SI with high-resolution target areas, and the low-resolution background in the reconstructions.