Blind Restoration of Motion Blurred Images Using Improved Generative Adversarial Networks

Abstract

The invention proposes a method for blind restoration of motion blurred images using an improved generative confrontation network, and obtains a clear image through the confrontation training of the generative network and the discriminant network. The generative network is a fully convolutional neural network combined with a residual network, which can deepen the network layer while reducing training time. The discriminant network is a binary classification network composed of convolutional layer, pooling layer and fully connected layer, which is used to determine the image restored by the generating network or the original clear image. The loss function adopts the minimum mean square error of smooth and unsaturated gradients, which can optimize network training and avoid gradient disappearance. At the same time, an image fidelity item is added to the loss function of the generation network to constrain the distribution of the restored image to be closer to the clear image.

Description

technical field [0001] The invention belongs to the technical field of image processing. Background technique [0002] Blind restoration of blurred images is a basic subject in image processing tasks. Restoring motion blurred images when the blur kernel is unknown is a serious ill-posed problem. In the traditional blind restoration of blurred images, the image and blur kernel are mainly constrained by various prior knowledge, and a strongly constrained image restoration model with generalization ability is established. Accurate blur kernel estimation is the key, and a fast model is required For this reason, these problems have become the bottleneck that affects the practical application of blurred image blind restoration technology for a long time. [0003] Deep learning has the characteristics of automatic feature extraction and powerful computing power, and has become the development direction of blurred image restoration research. Although deep learning can realize the ...

AI Technical Summary

Problems solved by technology

The problem with this type of method is that it is difficult to calculate and the process of tuning parameters is complicated. At the same time, it is still necessary to estimate the blur kernel in most cases.

Recently, some scholars have proposed a full network that does not estimate the fuzzy kernel, but the research is still in its infancy, the designed network is more complex, and the training is unstable

Images