Abstract:

Recently, the generative adversarial networks have shown good potential in the generation of constrained image, which makes it suitable for image super-resolution reconstruction. In view of the problem of low utilization of feature information in image super-resolution reconstruction algorithm based on convolution neural network, this paper proposes a super-resolution reconstruction algorithm for residual dense generative adversarial network on the basis of the framework of generative adversarial network. The algorithm defines the generator network and the discriminator network. Through constructing the residual dense network as the generator network and PatchGAN as the discriminator network, it solves the problem of the low utilization rate of feature information in the super-resolution algorithm based on convolution neural network and the problem of the slow convergence of the generative adversarial network. The reconstruction algorithm is compared with the mainstream super-resolution reconstruction algorithm on the basis of Set5 and other standard datasets. The experimental results show that the proposed algorithm can effectively improve the utilization rate of feature information, restore the details of low-resolution images, and improve the quality of image reconstruction.

Key words: super-resolution reconstruction, generative adversarial networks, residual dense network, PatchGAN

摘要：

近年来，生成对抗网络在约束图像生成方面表现出了较好的潜力，使其适用于图像超分辨率重建。针对基于卷积神经网络的图像超分辨率重建算法存在的特征信息利用率低的问题，基于生成对抗网络框架，提出了残差密集生成对抗网络的超分辨率重建算法。该算法定义生成器网络、判别器网络，通过构建残差密集网络作为生成器网络及PatchGAN作为判别器网络，以解决基于卷积神经网络的超分辨率算法中特征信息利用率低以及生成对抗网络收敛慢的问题。该重建算法在Set5等标准数据集上与主流的超分辨率重建算法进行对比，实验表明，该算法能够有效地提高特征信息利用率，较好地恢复低分辨率图像的细节信息，提高图像重建的质量。

关键词: 超分辨率重建, 生成对抗网络, 残差密集网络, PatchGAN