基于改进SRResNet深度学习网络的低照度图像超分辨率重建方法

为解决低照度条件下低分辨率图像的超分辨率重建问题，提出一种基于改进超分辨率残差网络(super-resolution residual networks, SRResNet)深度学习网络的低照度图像超分辨率重建方法。通过将读取的图像下采样、亮度降低等处理生成低照度低分辨率图像，并将该图像与高分辨率图像作为数据对输入学习模型，以便改进SRResNet的训练数据对的生成方式，优化训练过程，从而构建面向单帧低照度彩色图像的基于改进SRResNet训练的超分辨率重建方法。实验结果表明：与现有流行的图像超分辨率重建方法相比，该方法的峰值信噪比(peak signal to noise ratio, PSNR)、结构相似性(structural similarity, SSIM)指标整体为最优，低照度环境下的超分辨率重建图像更为清晰明亮、细节更丰富，该方法训练出的深度学习网络的重建效果更好。

Super-Resolution Reconstruction Method for Low Illumination Images Based on Improved SRResNet Deep Learning Network

In order to solve the problem of super-resolution reconstruction for low illumination images, a super-resolution reconstruction method based on improved Super-Resolution Residual Networks (SRResNet) is proposed for low illumination images with low resolution. In order to improve the generation method of training image pairs and optimize the training process, the low illumination image with low resolution is generated by down sampling and brightness reduction of the high resolution (HR) image, and the processed image and the corresponding HR image are input into the learning model as the data pair, so that a super-resolution reconstruction method for single frame low illumination color image based on improved SRResNet is constructed. The experimental results show that compared with the popular image super-resolution reconstruction methods, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) indicators of the proposed method are the best as a whole among all the comparison methods. This proposed method reconstructs clearer, brighter and more detailed super-resolution image from low illumination image with low resolution. The depth learning model trained by this method has better super-resolution reconstruction effect on low illumination image.