Blurred variable image blind restoration method based on deep convolution network

Abstract

The invention discloses a blurred variable image blind restoration method based on a deep convolution network. The method comprises the following steps: S1, types of blurs that need to be considered for blurred images are modeled, and a range of blurred parameters is determined for each type of blur; S2, according to the range of blurred parameters for each blurred type determined in step S1, a training data set with variable blur is synthesized; each group of training data comprises clear images, blurred images and blur parameters; S3, a deep convolutional network is built and trained, a training data set is input into a neural network, weight of the neural network is optimized, and a trained deep convolutional network is obtained; S4, blurred images to be restored are input into the trained deep convolution network, and reconstructed clear images are output. The method can be free from limitation of a conventional deblurring algorithm, a full convolutional network is adopted for estimating a plurality kinds of variable blur, and precision and speed of the algorithm can be improved.

Description

technical field [0001] The invention relates to the fields of digital image processing and machine learning, in particular to a blind restoration method for fuzzy variable images based on a deep convolutional network. Background technique [0002] When a camera or other optical imaging device is used to image an object, the image is likely to be blurred due to many factors. Insufficient lighting conditions and too long exposure time will cause the relative motion between the camera and the subject to be recorded in the final image, which is what we often call motion blur. In the field of remote sensing photography, atmospheric turbulence will affect the final imaging effect and cause blurring. Another common type of blur stems from inaccurate shooting focus. In most cases, we do not have the conditions to retake a clear image, so it is necessary to perform a deblurring operation on the blurred image to restore a clear image. [0003] The blurring process of an image is us...

AI Technical Summary

Problems solved by technology

Although the above method introduces DNN as a method for estimating kernel parameters and makes good use of the learning ability of DNN for a large amount of data, it still has the following limitations: 1) it does not break away from the traditional clear image restoration method, and the running speed is slow; 2) for The prediction of the fuzzy kernel needs to be designed for different types of networks, which is not universal; 3) For the variable fuzzy kernel, block processing and weighted fusion are required. The block accuracy affects the restoration effect and consumes additional running time

Images