Depth image deblurring method based on multi-scale fusion coding network

Abstract

The invention discloses a depth image deblurring method based on a multi-scale fusion coding network, and the method comprises the steps: carrying out the multi-scale coding fusion of an image; constructing a generative network model in cooperation with a region attention module and a feature fusion module; constructing a discriminant network model through a deep convolution module; generating an adversarial network model in combination with the generative network model and the discriminant network model; obtaining a clear and blurred image pair, inputting the clear and blurred image pair into the adversarial network model, continuously and alternately performing adversarial training to obtain an image deblurring modell stopping training after an error converges to a specified range; and performing image deblurring by using the image deblurring model to obtain a deblurred image. The technical problem of poor effect of a deep learning image deblurring model in the prior art is solved.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a method for deblurring deep images based on a multi-scale fusion coding network. Background technique [0002] Image deblurring, i.e. recovering sharp images from blurred images, has been an important research area in recent years. At present, the key is to recover the necessary edge details and the overall structure information of the image due to the motion blur of the captured image caused by the translation or rotation of the camera during the acquisition process. [0003] The traditional method uses the estimation of the fuzzy model, and uses different prior information (color, local smoothness, non-local self-similarity, sparsity, etc.) as a regularization term to improve the image deblurring effect. Most of these methods involve fixed parameters and a large number of calculation processes, and the actual blur is much larger than the estimated model, resulting in th...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a deep image deblurring method based on a multi-scale fusion coding network, aiming to solve the technical problem of poor effect of the deep learning image deblurring model in the prior art

Images