An end-to-end haze concentration adaptive neural network image dehazing method

Abstract

The invention discloses an end-to-end haze concentration adaptive neural network image dehazing method, comprising the following steps: constructing an image dehazing model; acquiring hazy image data; using a feature enhancement module in the image dehazing model to The images recovered from different paths are concatenated in the graph, and the blurred images with different dense haze are integrated together to help the network adaptively perceive the image haze concentration; the enhanced features are reconstructed into clear ones through the multi-scale feature attention module. haze-free image; calculate the mean square error and perceptual loss of the restored image and the corresponding clear image, and update the image dehazing model; the mean square error guides the image dehazing model to learn the content of the clear image, and the perceptual loss is used to quantify the restored image and the corresponding Visual differences between sharp images, two loss functions cooperate to optimize the dehazing model. The above technical solution effectively dehazes the actually photographed fog image, restores high-quality images, and has good practicability.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an end-to-end haze concentration adaptive neural network image dehazing method. Background technique [0002] With the development of technology, great breakthroughs have been made in some computer vision tasks such as object detection, object tracking, behavior analysis, and face recognition. However, high-level vision tasks such as detection and tracking rely on clear video and image data, and their performance is often greatly affected in real-world scenarios such as heavy fog and heavy rain. As a pre-task for some advanced vision tasks, image dehazing has attracted the attention of many researchers in recent years. [0003] Image dehazing is a typical image restoration problem, which can be traced back to 1924. McCartney et al. first proposed a classic atmospheric light scattering model, the model is as follows: [0004] I=tJ+A(1-t), [0005] t(x)=e βd(x) , [0...

AI Technical Summary

Problems solved by technology

However, different priors depend on the estimation of a certain feature of the image. In real scenes, these prior information are often invalid and computationally complex.

Images