Deep Coded Aperture Design: An End-to-End Approach for Computational Imaging Tasks

Covering from photography to depth and spectral estimation, diverse computational imaging (CI) applications benefit the versatile modulation of coded apertures (CAs). The lightwave fields as space, time, or can be modulated obtain projected encoded information at sensor that is then decoded by efficient methods, such modern deep learning decoders. Although CA fabricated produce an analog modulation, a binary preferred since more straightforward calibration, higher speed, lower storage are achieved. As performance decoder mainly depends on structure CA, several works optimize ensembles customizing regularizers for particular application without considering critical physical constraints CAs. This work presents end-to-end (E2E) learning-based optimization CAs CI tasks. design method aims cover wide range problems, easily changing loss function approach. designed includes fulfill widely used sensing requirements applications. Mainly, selected transmittance, compression ratio, correlation among measurements. At same solution encouraged, task maximized in restoration, classification, semantic segmentation.