Gaussian blur image restoration method based on generate antagonistic network

Abstract

The invention discloses a Gaussian blur image restoration method based on a generated antagonistic network, comprising two steps of training and learning the generated antagonistic network parametersand applying the generated antagonistic network to the Gaussian blur image restoration, and realizing end-to-end image restoration; generating countermeasure network consists of generator and discriminator. Generator and discriminator are mainly composed of convolution network layer. The training process of the network is: according to the constructed Gaussian fuzzy kernel set, the clear image isblurred, and the training set of clear image and blurred image data pairs is obtained, which is used to train and generate the antagonistic network. Given a single Gaussian blurred image, input the trained model, we can restore the clear image. By using the generated antagonistic network and learning and fitting ability of the convolution neural network, the invention can be further used for constructing a Gaussian blurred image restoration system to obtain a very good Gaussian blurred image restoration effect, and has important practical value in practice.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more specifically relates to a Gaussian blurred image blind restoration algorithm based on a generative confrontation network, which is used to remove Gaussian blurred images in images. Background technique [0002] Images are prone to quality degradation during acquisition. When the camera is in the high-speed airflow, the friction between the camera and the airflow will cause the airflow density to be uneven, making the image appear turbulent and blurred, and a Gaussian blurred image will be obtained. Image deblurring (also known as image restoration technology) is the technology of recovering a clear image from a blurred image. This is a very challenging and highly practical research content in image processing. Researchers mainly focus on using the existing image prior knowledge and propose many image deblurring techniques. Among them, according to whether the point spread fun...

AI Technical Summary

Problems solved by technology

Mainly focus on removing image Gaussian blur

At the same time, as a general model, it is also suitable for removing defocus blur and other problems.

Images