Hybrid Fourier light field microscopy system with deep learning for 3D high-resolution reconstruction

This study tackles the spatial resolution limitations of traditional light field microscopy by proposing a 3D resolution enhancement method that integrates innovations in optical system design and deep learning-based reconstruction. A hybrid Fourier light field microscopy (HFLFM) system is developed to capture high-resolution central views and multi-angle low-resolution light field images simultaneously, achieving an incompatible balance between image quality and angular information acquisition. Based on this system, a resolution enhancement network is constructed, featuring three key modules: a self-attention angular enhancement module for modeling inter-view consistency and global dependencies, a hybrid residual feature extraction module for enhancing high-frequency detail recovery, and a progressive resolution enhancement fusion module for fine-grained reconstruction. This architecture effectively constructs the geometric relationships and spatial details, improving both angular consistency and image clarity. Experiments on the dense light field dataset (DLFD), the HCI 4D light field dataset, and our hybrid microscopic system demonstrate superior performance in peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM), along with excellent 3D reconstruction accuracy. A lateral-resolution experiment verifies a fourfold improvement, and the max error in depth evaluation is reduced by approximately 88%, confirming the feasibility and precision, and of the proposed method and HFLFM system. Ablation studies further confirm the effectiveness of each module. In summary, the proposed method enhances the spatial resolution of light field microscopy through joint optimization of optical system innovation and associated network design for data processing, offering an effective and credible solution for high-quality 3D reconstruction without increasing system complexity.