Abstract:Correlation imaging indirectly reconstructs target images by leveraging the second-order or higher-order correlation properties of the light field, with reconstruction algorithms being critical to imaging performance. However, existing algorithms face challenges in balancing reconstruction capability and adaptability. To address this, a snapshot correlation imaging reconstruction algorithm based on the Plug-and-Play (PnP) framework and the Alternating Direction Method of Multipliers (ADMM) is proposed. By flexibly incorporating high-performance denoisers, the algorithm significantly improves reconstruction performance under both low and high sampling rates. Specifically, when using the deep convolutional neural network FFDNet as the denoiser, the method effectively combines the strengths of model-driven and learning-driven approaches, overcoming the limitations of traditional compressed sensing and deep learning methods. Experimental results demonstrate that this algorithm achieves high-quality image reconstruction, with both subjective and objective evaluation metrics outperforming commonly used compressed sensing algorithms.
