HSR-CFAT-based small-sample super-resolution reconstruction framework for off-axis digital holography

Digital holography is one of the key technologies for acquiring wavefront information of three-dimensional objects, and obtaining high-quality holograms is a prerequisite. To address the issue of low-resolution holograms caused by pixel size limitations and diffraction effects in imaging sensors (CCD/CMOS), the common approach involves using large datasets for super-resolution reconstruction through deep-learning methods. Due to hardware limitations and constraints in the experimental environment, it is more difficult to create large datasets. To address this challenge, this study proposes a small-sample super-resolution reconstruction framework based on holographic super-resolution CFAT (HSR-CFAT), for the first time, to the best of our knowledge. The HSR-CFAT model uses a residual hybrid attention group (RHAG) to fuse the hybrid attention block dense (HAB_D), hybrid attention block sparse (HAB_S), and overlapping cross-attention block (OCAB). Together with sliding window attention (SWA) and dynamic-resolution processing, it efficiently extracts multi-scale features to generate high-precision computer-generated holograms (CGHs). Further optimization of computational efficiency and phase accuracy through the phase-aware upsampling (PAU) module, achieving robust pixel-level reconstruction and overcoming the reliance of traditional methods on large datasets. A small-scale dataset was constructed by collecting 377 multi-view holograms. Super-resolution processing was performed directly on the holograms to overcome sensor pixel limitations, and the phase reconstruction module was fused to generate high-resolution phase images. The experimental results demonstrate that the network can process the collected multi-scale holographic images effectively, achieving significantly better performance than existing methods in terms of both subjective visual quality and objective evaluation metrics. This opens up a new technical pathway for super-resolution reconstruction in off-axis digital holographic imaging.