基于高斯模糊的CNN的单幅图像超分辨率重建算法

近几年卷积神经网络在单幅图像超分辨率重建工作中取得了很大的进步,但是大部分基于卷积神经网络(CNN)的单幅图像超分辨重建算法是建立在低分辨率图像由高分辨率图像通过双三次插值法下采样取得的前提下,当这个假设不成立时,图像重建的客观评价指标PSNR以及主观的视觉效果就会较差。针对此问题,提出一种基于高斯模糊的CNN的单幅图像超分辨率重建算法,通过在图像输入网络前,将原始低分辨率图像与高斯模糊核进行卷积,并进行低频信息融合以增强网络的泛化能力,使用亚像素卷积法把图像上采样到目标图像大小,进而消减网络的参数数量,提升运算速度。实验结果表明,该算法在不同放大倍数下的重建效果均优于传统算法。

SUPER-RESOLUTION RECONSTRUCTION ALGORITHM OF SINGLE IMAGE BASED ON CNN WITH GAUSSIAN BLUR

In recent years,convolutional neural network(CNN)has made great progress in single image super-resolution reconstruction.However,most of the single image super-resolution reconstruction algorithms based on CNN assume that the low-resolution image is obtained by sampling the high-resolution image through bicubic interpolation,so when the assumption does’t hold up,the objective evaluation index PSNR and subjective visual effect of image reconstruction may be very poor.To solve this problem,a CNN algorithm based on Gauss blur is proposed to reconstruct the super-resolution of a single image.Before the data was input into the network,the original low-resolution image was convoluted with Gauss blur kernel,and the low-frequency information was fused to enhance the generalization ability of the network.Secondly,the sub-pixel convolution method was used to sample the image to get the target image and reduce the number of network parameters,so as to improve the operation speed.The experimental results show that the reconstruction effect of the proposed method is better than that of the traditional algorithm under different magnification.