Incoherent light control through scattering media based on machine learning and its application to multiview stereo displays

Abstract

In this paper, we present a method for controlling incoherent light through scattering media based on machine learning and its potential application to multiview stereo displays. The inverse function between input and output light intensity patterns through a scattering medium is regressed with a machine learning algorithm. The inverse function is used for calculating an input pattern for generating a target output pattern through a scattering medium. We demonstrate the proposed method by assuming a potential application to multiview stereo displays. This concept enables us to use a diffuser as a parallax barrier, a cylindrical lens array, or a lens array on a conventional multiview stereo display, which will contribute to a low-cost, highly functional display. A neural network is trained with a large number of pairs of displayed random patterns and their parallax images at different observation points, and then a displayed image is calculated from arbitrary parallax images using the trained neural network. In the experimental demonstration, the scattering-based multiview stereo display was composed of a diffuser and a conventional liquid crystal display, and it reproduced different handwritten characters, which were captured by a stereo camera.