高斯滤波驱动的卷积神经网络单幅图像超分重建  

Single Image Super-resolution Reconstruction Based on Convolution Neural Network Driven by Gaussian Filter

作　　者：李慧娜 康雯婷 LI Huina;KANG Wenting(Journal Editorial Department,Xuchang University,Xuchang 461000,China;School of Science,Jilin Institute of Chemical Technology,Jilin 132000,China)

机构地区：[1]许昌学院学报编辑部,河南许昌461000 [2]吉林化工学院理学院,吉林吉林132000

出　　处：《许昌学院学报》2025年第2期111-116,共6页Journal of Xuchang University

基　　金：许昌学院科研专项(2023ZX045)。

摘　　要：为了改进单幅图像超分重建效果,把高斯滤波模块融合进重建流程中,提出高斯滤波驱动的卷积神经网络(CNN)单幅图像超分重建方法.高斯滤波模块先对受噪声干扰的图像进行降噪处理,降噪后的图像经过基于传统卷积神经网络创建的双分支超分重建模型完成分辨率提升.实验结果表明,新方法对受不同程度噪声污染的图像进行超分重建的效果要优于不引入高斯滤波模块的传统超分重建方法的效果,有较好的PSNR值和NIQE值.To enhance the performance of single image super-resolution reconstruction,the Gaussian filter is integrated into the reconstruction process,leading to the proposal of Gaussian filter-driven convolutional neural network(CNN)method for single-image super-resolution reconstruction.The Gaussian filter first performs denoising on images contaminated by noise,and the denoised images then undergo resolution enhancement through a CNN.Experimental results demonstrate that the new method,when applied to images polluted by varying degrees of noise,achieves superior super-resolution reconstruction compared to traditional methods that do not incorporate Gaussian filter.It exhibits better PSNR and NIQE values.

关 键 词：图像超分 高斯滤波 CNN PSNR NIQE 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]