Multidimensional Computational Imaging using Single-pixel Detectors

Abstract

espanolEl ser humano depende de su sentido de la vista para obtener informacion de sus alrededores. Sin embargo, este esta claramente limitado: somos incapaces de ver objetos muy lejanos o pequenos. Para afrontar estas limitaciones hemos desarrollado multitud de sistemas opticos. Sin embargo, hoy en dia aun quedan fronteras fisicas que nuestros dispositivos no son capaces de atravesar. En esta tesis veremos como es posible obtener imagenes con informacion sobre multiples dimensiones de la luz (polarizacion, fase, longitud de onda) utilizando sistemas basados en detectores de un solo pixel, moduladores espaciales de luz y tecnicas computacionales de procesado de senal. De este modo, es posible construir sistemas opticos que miden mayores cantidades de informacion que los sistemas tradicionales mientras se reduce el coste economico de los mismos y se aumenta su velocidad. EnglishHumans depend on vision to gather information about their surroundings. However, our sight sense is very limited: we cannot see very distant or small objects, and we can only sense light inside the visible spectra. To tackle these limitations, we have been developing optical sensing tools for more than four centuries now. However, even though nowadays we can even see objects at the nanometric scale, or very distant galaxies, there are still fundamental limitations that physical systems cannot bypass. In this thesis, I will show you how to obtain images with information about multiple dimensions of light (polarization, phase, wavelength) using novel sensing paradigms based on single-pixel detection and signal processing techniques. Using detectors without spatial resolution makes it possible to easily work in exotic spectral ranges, in low light level scenarios, or to build very compact and efficient multidimensional imaging systems. Moreover, the presence of a fast spatial light modulator in all of these systems allows to implement modern recovery techniques based on algorithmic approaches, such as compressive sensing or matrix completion. In doing so, these computational imaging systems can obtain more information than a traditional system, but in a faster and inexpensive manner.