Random scattering media that scatters incident light waves for optical diffraction-limit break imaging has important application prospects and significance in biomedical imaging, lithography micromachining and nanomaterial surface shape analysis. However, the transmission matrix, which represents the relationship between incident and output pixels in monochromatic image detecting through scattering media, suffers from severe noise and limited resolution in the reconstructed image owing to its imaging characteristics. In this study, we propose a deep convolution neural network to achieve single image enhancement effects, including denoising and super-resolution on images that are reconstructed by a monochromatic transmission matrix. We demonstrate significantly higher quantitative scores in peak signal-to-noise, SSIM and correlation coefficient compared to conventional methods. We further show that our method is effective for the transmission matrix-based reconstruction operator for both phase conjugation and pseudo-inversion.

散射入射光用于光学衍射极限断裂成像的随机散射介质在生物医学成像，光刻微加工和纳米材料表面形状分析中具有重要的应用前景和意义。然而，由于其成像特性，代表通过散射介质检测单色图像中的入射像素和输出像素之间的关系的传输矩阵遭受严重的噪声和有限的分辨率限制。在这项研究中，我们提出了一种深度卷积神经网络，以实现单幅图像增强效果，包括对由单色传输矩阵重构的图像进行去噪和超分辨率。我们证明了峰值信噪比的定量得分明显更高，与传统方法相比，SSIM和相关系数更高。我们进一步表明，我们的方法对于基于相位变换和伪反演的基于传输矩阵的重构算子都是有效的。