Search

Image deblurring aims to restore the latent sharp image from the blurred one. In recent years, some learning-based image deblurring methods have achieved significant advances. However, the tradeoff between the texture details and model parameters is still a crucial issue. In this paper, we propose a novel deblurring method based on two-level wavelet-based convolutional neural network (CNN), which embeds discrete wavelet transform (DWT) to separate the image context and texture information and reduces the complexity of calculation. Furthermore, we modify the Inception module by adding pixels-wise attention (PA) mechanism and channel scaling factor to make each convolution kernel have different weights, which increase the receptive field while significantly reduce the parameters of the module. Qualitative and quantitative evaluation on real-word and synthetic datasets shows that the deblurring performance of our method is comparable to the state-of-the-art algorithms. Moreover, compared to the traditional learning-based deblurring method, our model has fewer parameters.

Recently, deep convolutional learning has been applied to image deblurring, which greatly improves the performance of single-image blind deblurring algorithms. However, most deep image deblurring models based on convolutional neural networks do not make full use of the information of low-level spatial convolutions, so they achieve relatively poor performance. In this paper, we improved the generator of the Generative Adversarial Network (GAN) and proposed dense networks and residual networks to solve this problem in image deblurring. The network makes full use of the hierarchical features of all convolutional layers. Specifically, we extract the local features of the convolutional layer through the proposed densely connected convolutional layer. Moreover, the dense connections of the residual dense network can connect all layers from the previous state to the current state. Then using local jump connection and feature fusion, it can adaptively learn more effective features from the previous and current local features, making the trained network more extensive. After fully obtaining the dense local features, the global residual connection is used to learn the global feature information in an overall manner. Experimental comparison and verification have proved the effectiveness of our method and achieved excellent results both quantitatively and qualitatively.

Single image deblurring aims to retore the latent sharp images from the corresponding blurred ones, which is a highly ill-posed problem. In this paper, we present a deep recursive network based on generatetive adversarial networks (GANs) that restores sharp images in an end-to-end manner where blur is caused by various sources, our recursive neural network can greatly reduce the computation and complexity of the model. Then we use the least square discriminator to prevent the gradient from disappearing and make the training process more stable. We also add an adversarial loss to make the generated images look more realistic and a perceptual loss to generated better image. Experimental results have shown that our proposed method produces better performance and faster time.