Enhanced Compressive Imaging Using Model-Based Acquisition: Smarter sampling by incorporating domain knowledge

Abstract

Compressive imaging (CI) is a subset of computational photography where a scene is captured via a series of optical, transform-based modulations before being recorded at the detector. However, unlike previous transform imagers, compressive sensors take advantage of the inherent sparsity in the image and use specialized algorithms to reconstruct a high-resolution image with far lower than 100% of the total measurements. Initial CI systems exploited the properties of random matrices used in other areas of compressive sensing (CS); however, in the case of imaging, there are immense benefits to be derived by designing measurement matrices that optimize specific objectives and enable novel capabilities. In this article, we survey recent results on measurement matrix designs that provide the ability of real-time previews, signature-selective imaging, and reconstruction-free inference.