|
Imaging though scattering or strongly diffuse media is an outstanding challenge that persists despite significant progress over the years. We can characterise the scattering strength of a material by quantifying its thickness in terms of the number of transport mean free paths (TMFPs), which represents the distance over which all information of the photon’s initial direction is lost. This value can vary across various orders of magnitude depending on the medium, for example between 1 mm in biological tissue to 1 m in fog. A relatively straightforward approach to imaging in diffuse media is to time gate out all photons except the ``ballistic” photons that will have experienced little or no scattering. Unfortunately, ballistic photons are attenuated exponentially: imaging through 10 TMFPs implies a loss of 1e-5 that still allows detection with a decent laser and detector; 100 TMFPs implies an attenuation of roughly 30 orders of magnitude. All photons have therefore undergone significant scattering with the implication that image information must be spread over the full temporal distribution of the photon arrival times at the receiver. We will overview recent work aimed at retrieving image information from highly scattering media >>10 TMFPs) by using single photon sensitive cameras to record the full spatial and temporal distribution of transmitted photons, to which we then apply computational methods in order to retrieve images of embedded objects with sub-mm resolution. We will also discuss ongoing efforts to image through dynamically scattering media, i.e. that change in time, using machine learning approaches. We are able to successfully image objects even in situations in which the scattering is so strong that no speckle memory effect is present and the continuously changing medium does not allow to measure a transmission matrix, thus defeating all previous approaches to this problem.
|
