Raindrop removing method for single image based on dense multi-scale generative adversarial network

Abstract

The invention discloses a raindrop removing method for a single image based on a dense multi-scale generative adversarial network. Constructing a multi-scale image restoration model by using a dense network for feature reuse; constructing a discriminant network model with an attention mechanism by combining and utilizing the multi-scale image restoration model; forming a multi-scale generative adversarial network model; obtaining an original rain image, an original rain-free image and a residual raindrop layer; inputting the original rain-free image and the residual raindrop layer into the discrimination network model; utilizing an error between the discriminant network model and the generative network model; and performing back propagation to alternately train the multi-scale generative adversarial network model, stopping training until errors of the discrimination network model and the generative network model converge to a set range, generating a raindrop removal model by using thetrained generative network model, and removing relatively large and dense raindrops in a single image.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to a method for removing raindrops from a single image based on a dense multi-scale generation confrontation network. Background technique [0002] In many applications such as drone-based video surveillance and autonomous vehicles, raindrops adhering to glass windows, windshields, or lenses can obstruct the visibility of background scenes and degrade image quality. Mainly because the image contained in the raindrop area is different from the image without the raindrop area, and, in most cases, the focus of the camera is on the background scene, making the appearance of the raindrop blurred, and some methods have been proposed to solve the problem of raindrop detection. and removal problem, dedicated to detecting raindrops but not removing them, other methods were introduced to detect and remove raindrops using stereo, video, or specially designed optical shutters, ...

AI Technical Summary

Problems solved by technology

Mainly because the image contained in the raindrop area is different from the image without the raindrop area, and, in most cases, the focus of the camera is on the background scene, making the appearance of the raindrop blurred, and some methods have been proposed to solve the problem of raindrop detection. and removal problem, dedicated to detecting raindrops, but not removing them, other methods were introduced to detect and remove raindrops using stereo, video, or specially designed optical shutters, but not suitable for a single input image taken by a common camera

In the method of deep learning, there is a method of removing raindrops or dust from an image, but it can only deal with small raindrops, and the effect is not good, and it cannot deal with relatively large and dense raindrops.

Images