基于改进残差亚像素卷积神经网络的超分辨率图像重建方法研究

Super-resolution Image Reconstruction Based on Improved Residual Sub-pixel Convolution Neural Network

为了提高重建图像的分辨率,降低算法的复杂性,保留更多的边缘信息,提出一种改进的残差亚像素卷积神经网络超分辨率图像重建方法。首先,在网络输入与输出之间增加一个新的通道,使卷积网络只学习LR与HR的残差,减少网络冗余。其次,在输出层应用亚像素卷积网络进行上采样,对特征图像重新排列,得到最终的高分辨率图像。实验结果表明,本文方法在分辨率效果、图像纹理、视觉效果等方面均优于经典方法(SRCNN方法和双三次插值方法)。在Set5、Set14、BSD100和DIV2K等常用数据集上的峰值信噪比(PSNR)和结构相似比(SSIM),与SRCNN方法相比都有提高,同时改进了图像重建算法的性能。

An improved super-resolution image reconstruction method is proposed based on residual sub-pixel convolution neural network in order to improve the resolution of the reconstructed image,and reduce the complexity of the algorithm and retain more edge information.First,a new channel is added between the input and output of the network,so that the convolution network can only learn the residual of LR and HR and reduces the network redundancy.Secondly,in the output layer,the sub-pixel convolution network is used for sampling and the feature image is rearranged to get the final high-resolution image.Experiment results show that this method has achieved better super-resolution effect compared with the classical methods such as the SRCNN method and the Bicubic interpolation method.At the same time,the Peak Signal to Noise Ratio(PSNR)and the Structural Similarity Index(SSIM)is improved in common data sets such as Set5,Set14,BSD100 and DIV2K compared with SRCNN,which improves the performance of image reconstruction algorithm.