Physically-based phasor imaging for advanced scene understanding

Abstract

General imaging systems acquire color information of light. With the development of sensor technology, the modern camera can capture a high resolution similar to that of the human visual system. However, to analyze the characteristics of an object, color information is insufficient, and it is necessary to acquire the spectrum and geometric information of the object. With advances in computational photography, spectral information and depth information can be acquired with high quality. However, many imaging systems have disadvantages regarding complex optical systems and large sizes. This thesis introduces imaging system hardware and computational photography algorithms using phasor analysis that can acquire high-quality 3D and spectral information without using such a complex optical system. First, a compact hyperspectral camera using a diffractive optical element optimized by phasor analysis is proposed. In conventional hyperspectral cameras, each optical element is used for a specific role, such as an imaging lens for taking a picture, a spectroscopic system for acquiring spectral information, and a relay lens for connecting them. As a result, such a system has no choice but to increase in size because a large number of optical elements are used. The proposed diffractive optical element can make a hyperspectral imaging system compact because a single diffractive optical element performs both imaging and spectrometer roles. In addition, high-quality hyperspectral images were obtained using the optimization algorithm-based reconstruction depth learning network. The second is high-fidelity timeof-flight (ToF) imaging in scattering media. ToF camera captures depth information by measuring the flight time of light rays. However, rays are randomly reflected by scattering media which causes multipath interference (MPI). We develop a robust depth acquisition method in scattering media by analyzing summed phasor by MPI. Using a linear polarization filter, we capture polarized light and de-polarized light of backscattered light and reconstruct depth by phasor difference of the two components.