Unfolding Framework with Complex-Valued Deformable Attention for High-Quality Computer-Generated Hologram Generation

2025-08-29

AI Summary

Conventional deep learning–based computer-generated holography (CGH) suffers from poor physical interpretability, limited convolutional receptive fields, and near-field constraints inherent to the angular spectrum method (ASM). Method: This paper proposes a physics-inspired deep-unfolding framework that (1) decouples gradient descent into an adaptive bandwidth-preserving module and a complex-valued phase-domain denoiser to enhance physical consistency and flexibility; and (2) introduces a complex-valued deformable self-attention mechanism to overcome local receptive field limitations and model long-range wavefront dependencies. Contribution/Results: The framework enables high-fidelity CGH generation over wide working distances and with global contextual awareness. Quantitative evaluation shows peak signal-to-noise ratios (PSNR) exceeding 35 dB on both simulated and experimental datasets—significantly outperforming state-of-the-art methods—while ensuring high accuracy, robustness, and physical interpretability.

Technology Category

• Computer Vision: Computational Photography, Image & Video Synthesis
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Computer-generated holography (CGH) has gained wide attention with deep learning-based algorithms. However, due to its nonlinear and ill-posed nature, challenges remain in achieving accurate and stable reconstruction. Specifically, ($i$) the widely used end-to-end networks treat the reconstruction model as a black box, ignoring underlying physical relationships, which reduces interpretability and flexibility. ($ii$) CNN-based CGH algorithms have limited receptive fields, hindering their ability to capture long-range dependencies and global context. ($iii$) Angular spectrum method (ASM)-based models are constrained to finite near-fields.In this paper, we propose a Deep Unfolding Network (DUN) that decomposes gradient descent into two modules: an adaptive bandwidth-preserving model (ABPM) and a phase-domain complex-valued denoiser (PCD), providing more flexibility. ABPM allows for wider working distances compared to ASM-based methods. At the same time, PCD leverages its complex-valued deformable self-attention module to capture global features and enhance performance, achieving a PSNR over 35 dB. Experiments on simulated and real data show state-of-the-art results.

Problem

• Addresses inaccurate and unstable hologram reconstruction
• Overcomes limited receptive fields in CNN-based holography
• Extends working distance beyond finite near-field constraints

Innovation

• Deep Unfolding Network with gradient descent modules
• Adaptive bandwidth-preserving model for wider distances
• Complex-valued deformable attention capturing global features