Self-evolving ghost imaging

Abstract

Ghost imaging captures 2D images with a point detector instead of an array sensor. It could therefore solve the challenge of building cameras in wave bands where sensors are difficult and expensive to produce and could open up more routine THz, near-infrared, lifetime, and hyperspectral imaging simply by using single-pixel detectors. Traditionally, ghost imaging retrieves the image of an object offline by correlating measured light intensities with pre-designed illuminating patterns. Here we present a “self-evolving” ghost imaging (SEGI) strategy for imaging objects bypassing offline post-processing. It also offers the capability to image objects in turbid media. By inspecting the optical feedback, we evaluate the illumination patterns by a cost function and generate offspring illumination patterns that mimic the object’s image, bypassing the reconstruction process. At the initial evolving state, the object’s “genetic information” is stored in the patterns. At the following imaging stage, the object’s image ( 48 × 48 p i x e l s ) can be updated at a 40 Hz imaging rate. We numerically and experimentally demonstrate this concept for static and moving objects. The frame-memory effect between the self-evolving illumination patterns provided by the genetic algorithm enables SEGI imaging through turbid media. We further demonstrate this capability by imaging an object placed in a container filled with water and sand. SEGI shows robust and superior imaging power compared with traditional computational ghost imaging. This strategy could enhance ghost imaging in applications such as remote sensing, imaging through scattering media, and low-irradiative biological imaging.