Machine learning accelerated holographic near-eye display system based on three-step diffraction

Holographic displays enable the formation of a three-dimensional particle field distribution and is promised to be one solution of the key problems in present virtual and augmented reality displays. The biggest challenge in the computer-generated holography (CGH) is the high consuming calculation time and the restricted field of view due to the micro display screen. In order to solve the above mentioned problems, we propose a method for machine learning accelerated lensless holographic projection, which is based on the reconstruction of complex amplitude image from phase-only CGH. The runtime of generating the phase-only CGH is half of traditional iterative method. The complex amplitude image can achieve the maximum diffraction bandwidth of the spatial light modulator (SLM). The feasibility of this method is verified by experiments.