基于ResNeXt和WGAN网络的单图像超分辨率重建

Single image super resolution based on ResNeXt and WGAN

为解决现有基于生成对抗网络的单图像超分辨率重建模型SRGAN网络训练不稳定、学习速率慢等问题,提出了一种基于ResNeXt和WGAN的单图像超分辨率重建模型Res_SRGAN。该模型参考ResNeXt网络结构构造生成器,降低了生成器的复杂度,仅为SRGAN的1/8;通过WGAN来构造判别器解决了SRGAN模型不稳定的问题;实验结果表明,在四个公开数据集上所提模型相较于现有单图像超分辨率重建模型在主客观评价中均取得了更加优越的性能。

To solve the problem of unstable training and slow learning speed problems of a generative adversarial network for image super-resolution(SRGAN),the paper proposed a single image super-resolution reconstruction model called the Res_SRGAN based on ResNeXt and WGAN.The model referred to ResNeXt network structure construction generator,which reduced the computational complexity of model generator to 1/8 that of the SRGAN.The discriminator was constructed by WGAN,which solved SRGAN’s instability.Experimental results demonstrate that the proposed model achieves better performance in subjective and objective evaluations using four public data sets compared with other single-image super-resolution reconstruction models.