Generative adversarial network image deblurring algorithm based on RRDB

The invention discloses a generative adversarial network image deblurring algorithm based on RRDB. On the basis of a deblurring model DeblurGAN network, an RRDB network unit integrating a multi-layer residual network and dense connection is used for replacing an RB network unit in the generator, and global jump connection is added between the input and the output of the generator, so that the learning and generation capabilities of the generator are improved. Then, in a loss function, the Waserstein distance is used as the confrontation loss of the network, and the problem of network degradation in the training process is avoided; and a pixel space loss function is added to the content loss, and the pixel content consistency of the generated image is constrained to finally generate a clear image. 本发明公开了一种基于RRDB的生成对抗网络图像去模糊算法。本发明在去模糊模型DeblurGAN网络的基础上，使用融合了多层残差网络和密集连接的RRDB网络单元替换生成器中的RB网络单元，并且在生成器的输入和输出之间加入全局跳跃连接，来提高生成器的学习和生成能力。然后在损失函数中，使用Wasserstein距离作为网络的对抗损失，避免在训练过程中出现网络退化问题；在内容损失上加入像素空间损失函数，对生成图像的像素内容一致性进行约束