Mask-based coded imaging systems and image reconstruction algorithms

Abstract

Computational imaging is an emerging field. Its rapid development has drawn tremendous attention from both research and commercial points of view. Unlike traditional imaging, which separately considers the optical imaging and computational processing, computational imaging combines the power of the optical elements and signal processing techniques to achieve augmented capabilities.

Previous work on various aspects of computational imaging has shown the powerful abilities that computations can bring into the imaging systems. However, the research is still in an early stage. Some drawbacks need to be conquered. For example, in compressed sensing (CS) related systems, the reconstruction quality cannot be satisfactory due to the ill-posed nature of the problem. Likely, in computational photography, the systems share a major defect. That is, as four-dimensional radiance information is recorded by a regular two-dimensional sensor, an unavoidable sacrifice of the spatial resolution has to be made to resolve angular differences. This eventually causes the low spatial resolution output.

To meet these challenges, more efforts have to be made in both imaging part and computational part. In this dissertation, we concentrate ourselves on a more specific form of computational imaging, i.e., mask-based coded imaging systems. In particular, the first part of the dissertation focuses on a mask-based terahertz (THz) CS imaging system. There we focus on the computational part and explore the reconstruction algorithms that can estimate the underlying scene as accurately as possible. After that, we discuss the lightfield photography and show that by combining the system modification and proper postprocessing algorithms, we can achieve a high-resolution lightfield. The corresponding simulation demonstrates the performance of our methods.