Speckle-based image reconstruction via uniform illumination in single multimode fiber endoscopic imaging systems

Abstract

Conventional multimode fiber (MMF) imaging systems based on Gaussian illumination often suffer from uneven speckle distributions, which limit image reconstruction quality and cause significant performance degradation under perturbations such as fiber deformation or temperature fluctuations. To overcome these challenges, this study proposes a uniform illumination scheme enabled by beam shaping, where a single MMF simultaneously serves as both the illumination unit and imaging probe. The input Gaussian beam is transformed into a uniform beam to generate more evenly distributed illumination speckles. In parallel, a lightweight U-shaped neural network architecture (USNet) is developed to enable efficient image reconstruction. Experimental results show that under static conditions, the combination of uniform illumination and USNet achieves SSIM values of 0.8105, 0.7056, and 0.8137 on the MNIST, Fashion-MNIST, and SIPaKMeD datasets, respectively. Under thermal perturbation, the uniform speckle pattern significantly enhances system stability; although its robustness under mechanical disturbance is slightly lower than Gaussian illumination, the proposed method still maintains superior reconstruction performance within a practical perturbation range. Furthermore, additional evaluations under simulated subjective speckle conditions and varying MMF lengths demonstrate the method’s strong scalability and clinical application potential. This study is the first to systematically validate the robustness advantages of uniform illumination in MMF imaging and integrates a lightweight network to achieve high-quality image reconstruction, offering a new pathway and theoretical foundation for dynamic minimally invasive medical imaging with MMFs.