Single-pixel imaging with machine learning and compressed sampling methods

Single pixel imaging is an imaging technique that uses a single photodetector instead of a traditional multi-pixel matrix sensor. The key feature of the method is the use of spatial modulation of light (a sequence of light patterns). Instead of capturing an image with a multi-pixel sensor (as a regular camera does), a single-pixel camera forms an image by measuring the integral overlap of an object and a light pattern, and the image is reconstructed from a set of such numbers using special computational algorithms. This method allows to obtain images in modes that are unattainable or difficult to achieve with standard cameras. For example, it becomes possible to perform single-photon detection, detect in spectral ranges outside the sensitivity range of silicon (infrared, terahertz, gamma), and obtain images of objects outside the line of sight or in highly scattering media.
In the talk, I discuss methods for computing images in single-pixel imaging based on compressed sampling and machine learning, as well as their experimental implementation.