РЕКОНСТРУКЦИЯ ИЗОБРАЖЕНИЙ В ДИФРАКЦИОННО-ОПТИЧЕСКИХ СИСТЕМАХ НА ОСНОВЕ СВЕРТОЧНЫХ НЕЙРОННЫХ СЕТЕЙ И ОБРАТНОЙ СВЕРТКИ

In recent years, several pioneering works were dedicated to imaging systems based on simple diffractive structures like Fresnel lenses or  phase zone plates. Such systems are much lighter and cheaper than  classical refractive optical systems. However, the quality of images  obtained by diffractive optics suffers from stronger distortions of  various types. In this paper, we show that a combination of the high- precision lens design with post-capture computational reconstruction allows one to attain a much higher image quality. The proposed reconstruction procedure uses a sequence of color correction, deconvolution, and a feedforward deep learning neural  network. An improvement both in lens manufacturing and in image  processing may contribute to the emergence of ultra-lightweight  imaging systems varying from cameras for nano- and picosatellites  to surveillance systems.