Efficient high-resolution microscopic ghost imaging via sequenced speckle illumination and deep learning from a single noisy image

This study presents a novel approach for achieving high-quality and large-scale microscopic ghost imaging by integrating deep learning-based denoising with computational ghost imaging techniques. By utilizing sequenced random speckle patterns of optimized sizes, we reconstructed large noisy images with fewer patterns while successfully resolving fine details as small as 2.2 μm on a USAF resolution target. To enhance image quality, we incorporated the Deep Neural Network-based Noise2Void (N2V) model, which effectively denoises ghost images without requiring a reference image or a large dataset. By applying the N2V model to a single noisy ghost image, we achieved significant noise reduction, leading to high-resolution and high-quality reconstructions with low computational resources. This method resulted in an average Structural Similarity Index (SSIM) improvement of over 324% and a resolution enhancement exceeding 33% across various target images. The proposed approach proves highly effective in enhancing the clarity and structural integrity of even very low-quality ghost images, paving the way for more efficient and practical implementations of ghost imaging in microscopic applications.