Noise Suppression on In-line Holograms with a Hybrid Method Using U-Net and Transformer Architectures

Digital holography enables capturing three-dimensional images from imaging methods but is often hindered by significant noise and interference artifacts. In this work, we propose a deep learning-based denoising method for inline holograms that leverages a hybrid CNN-Transformer U-Net architecture with multi-step Transformer refinement. Our approach integrates convolutional neural networks for local feature extraction with Transformer-based self-attention for modeling global context. In the experimental results, the performance of the proposed method was evaluated using PSNR, SSIM, and MSE metrics, and promising results were obtained.