基于特征补偿的深度神经网络重建超分辨率图像

Super-resolution Image Reconstruction Based on Feature Compensation Depth Neural Network

为有效恢复图像的高频信息,本文提出一种基于特征补偿的深度神经网络重建超分辨率图像方法.该方法结合密集型深度卷积神经网络和残差网络,并将原图像的高频信息单独提取上采样后与重建后的超分辨率图像融合形成高频特征补偿,使得图像质量提升.通过实验对比,本文算法相比于SRCNN算法重建出的超分辨率图像效果提升约1 db.

In order to effectively restore the high-frequency information of images, a depth neural network image super-resolution reconstruction method based on feature compensation is proposed. This method combines the dense depth convolution neural network and residual network, extracts the high-frequency information of the original images separately and fuses it with the reconstructed super-resolution images to form high-frequency feature compensation, which improves the image quality. Compared with SRCNN, the super-resolution image reconstructed by SRCNN is improved by about 1 db.