Fast Digital Hologram Generation Based on Parallel Processing

Abstract

Three-dimensional (3D) movies and television (TV) programs are becoming popular, and demand on 3D display technology is increasing. The principle of 3D display can be classified into two types, binocular disparity based stereoscopy and holography. Stereoscopy can make viewers feel fatigued when watching 3D video for a long period, whereas holography can display the 3D image as if a viewer sees the real scene with their naked eyes, avoiding fatigue. Hence, holography has been receiving considerable attention as the preferred 3D display technology. Digital holography can generate holograms without an optical system using only a personal computer (PC). Although the practicality of the technique is very high, it requires a huge amount of computation. Therefore, fast digital hologram generation is an important issue. This dissertation proposes fast digital hologram generation using parallel processing. The proposed methods are implemented by a general purpose graphic processing unit (GPGPU), and can be classified into single GPU, multiple GPU, and PC cluster based methods. The PC cluster system comprises multiple PCs with fast computer generated hologram calculation system utilizing multiple GPUs. The system quickly generates a high quality hologram using the PC cluster, allowing hologram resolution up to 1,536 1,536 pixels and two million light sources. In experimental tests implementing the proposed method, a single GPU based system generated a digital hologram with 1,920 1,080 pixel resolution and 1,000 light sources in 32.9 ms, approximately 2,400 times faster than a CPU based system; a multiple GPU based system generated a digital hologram with 1,920 1,080 pixel resolution and 1,000 light sources in 17.2 ms, approximately 10,000 times faster than a CPU based system; a PC cluster based system, composed of a server PC with nine client PCs, generated a digital hologram with 1,536 1,536 pixel resolution and 2,155,898 light sources in 10.3 seconds, approximately 4.7 times faster than a single PC based system. To verify the practicality of the proposed method, an application was developed to generate of digital mixed reality video holograms. Experimental comparisons showed the proposed application naturally mixes depth images of real and virtual objects in free viewing angles of a color depth camera, and solves the occlusion problem using Z buffers. Furthermore, the application executed all processes, including capturing the color and depth image, mixing depth images of real and virtual objects, and generating a digital mixed reality hologram, at 7.6 fps. Therefore, the proposed method can efficiently generate digital mixed reality video holograms.