| At present,information technology is developing rapidly and the popularity of shooting devices such as smart phones and DSLR cameras is becoming more and more widespread.As one of the most important information carriers in the network,images can hold rich information and help people record the moments in their lives,which are widely used in our daily life.Image deblurring,as a part of the image restoration field,is also a research hotspot in the field of computer vision.It has great research and application value.With the continuous development of artificial intelligence,deep learning plays an important role in the image field.Generative adversarial networks have a wide application in the field of image deblurring.How to improve the quality of recovered images,improve the network performance and generalization ability in real scenes are the issues that need to be focused on.In this proposal,we improve and optimize the generative adversarial network.We use Convolutional Neural Network as the basic network of generator and discriminator,use ResNetv2 instead of ResNet residual network module in the generator,and add global residual connections in the generator design,and use Instance Normalization to remove inter-sample dependencies.The design of the discriminator is borrowed from PatchGAN and WGAN.to obtain good image recovery results.The model in this work is trained using the GOPRO dataset and tested on the Kohler dataset and Lai dataset.The experiments show that satisfactory deblurring results can be achieved on the GOPRO dataset and Lai dataset.The experimental results are better than the existing methods when the three objective evaluation methods of Peak Signal To Noise Ratio,Structural Similarity Index,and Object detection method are evaluated together.The image deblurring effect can be effectively improved. |
