Presentation Information
[A-15-10]NEURAL NETWORK-BASED COMPLEX DATA COMPRESSION FOR COMPUTER-GENERATED HOLOGRAPHY
〇Norifumi Kawabata1 (1. Kanazawa University)
Keywords:
Computer-Generated Holography,Complex Data Compression,Neural Network,Image Quality Assessment,Cross Reality
Expectations for cross-reality (XR) technologies have increased with the metaverse's advancement. The author has evaluated the impact of coding quality on user immersion using aerial display prototypes. For computer-generated holography (CGH), efficient compression of vast complex wavefront data is indispensable. However, existing studies rarely evaluate the correlation between artifacts and human visual characteristics. This study proposes an evaluation framework to clarify the impact of neural compression on holographic visual quality. As a first step, a simulation using 2D-FFT and 8-level complex quantization was conducted on five images, including natural, 3DCG, and an ultra-high-resolution omnidirectional image captured with a RICOH THETA Z1. Objective evaluations (PSNR, SSIM) showed that quality degraded significantly for natural and omnidirectional images, whereas a specific 3DCG image exhibited higher tolerance. Results demonstrate that degradation behavior varies greatly depending on image content and edge sharpness. Future work will implement neural pipelines and quantify the correlation between degradation and subjective perceptual limits using LPIPS and user tests.