Image defogging method based on weak supervision generation antagonism network

Abstract

The invention discloses an image defogging method based on weak supervision generation antagonism network. The present invention mainly solves the problem of instability of defogging effect of the real foggy day image in the prior art. The implementation scheme is as follows: 1) obtaining a real training set and a composite training set; 2) respectively constructing a generation network and a confrontation network; 3) updating the parameters of the countermeasure network according to the loss function of the countermeasure network; 4) establishing a loss function of the generating network andupdating parameters of the generating network according to the loss function; 5) judging whether the update times of the generation network and the antagonism network reach 100, if so, inputting the real foggy day image into the generation network for defogging, otherwise, returning to step 3). The image after the method of the invention has rich details, improves the brightness, contrast and clarity of the image, reduces the phenomenon of supersaturation and distortion, and can be used in the field of computer vision.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to an image defogging method, which can be used for image preprocessing in computer vision tasks such as video monitoring, target detection and recognition. Background technique [0002] With the advent of the era of artificial intelligence, computer vision systems have been widely used in various fields such as public security, intelligent transportation, aerospace and satellite remote sensing, providing convenience and security for people's life and work. High-quality and clear images are the basic prerequisite for computer vision systems to work effectively, but most computer vision systems are very sensitive to weather conditions and light changes. In hazy weather, various particles suspended in the atmosphere not only absorb and scatter the incident light, but also scatter the surrounding ambient light and participate in the imaging process, which reduces the dynamic ran...

AI Technical Summary

Problems solved by technology

The degradation of foggy images directly affects the normal work of existing outdoor imaging systems, causing great inconvenience to subsequent computer vision tasks such as image segmentation, target recognition and tracking, and behavior detection, and bringing huge safety hazards to people's lives , so the research on image defogging methods is of great significance

[0003] At present, there are three main methods of image defogging: the first one is based on the enhancement method, which directly uses the existing image enhancement method to improve the contrast and clarity of the image. This method can effectively highlight the image details, but it is prone to oversaturation. Phenomenon, can not achieve the purpose of fog removal from the root

The second is a method based on a physical model. This method establishes an atmospheric scattering model based on the degradation of the foggy image, and combines the prior knowledge of the image to solve the parameters in the model, and then reversely deduces the fog-free image. This method realizes the true meaning of The image above is defogged, but it is not robust to scene changes

The third is a learning-based method, which artificially synthesizes foggy images through an atmospheric scattering model, uses machine learning or deep learning methods to estimate the transmittance of foggy images, and then reversely deduces fog-free images. This method can effectively extract The characteristics of foggy images, but relying on a large number of labeled synthetic image datasets, the dehazing effect on real foggy images is not ideal

The disadvantages of this method are: additional estimation of atmospheric light is required, parameters cannot be jointly optimized, and it only relies on labeled synthetic datasets, without using real foggy image datasets, the brightness of the image after dehazing is low, and the contrast is low. And the definition is not high enough, the visual effect is not ideal enough

Images