A Non-Paired Image Contrastive Dehazing Method Based on Feature Decoupling

Abstract

The invention discloses a feature-decoupling non-pair image contrast defogging method, which includes extracting features and performing comparative learning, which can effectively realize image defogging in non-pair scenes. In the absence of paired foggy and clear outdoor images in the same scene as training samples, the method of the present invention can fully consider unidirectional image generation, and use the self-supervision information of the image itself to effectively improve non-paired images Dehazing effect, a feature decoupling-based contrastive dehazing method is proposed to solve the non-paired dehazing problem, which can effectively achieve image dehazing in non-paired scenes.

Description

technical field [0001] The present invention relates to the technical field of image defogging, in particular to a non-paired image comparison defogging method based on feature decoupling, which can realize image defogging under non-paired foggy scene data sets. Background technique [0002] In foggy conditions, due to the scattering of atmospheric aerosol particles, the visibility of the image is severely reduced. In addition to causing a decrease in the visual perception of the image, the fog factor will also affect the accuracy of advanced tasks such as detection and segmentation. , which greatly limits the application value of the image. In addition, in the real world, it is difficult for us to obtain pairs of foggy and clear images. Due to the lack of paired foggy and clear outdoor images in the same scene as training samples, the ability of learning-based methods to remove fog is limited. Therefore, unpaired single image dehazing is very important in computer vision. ...

AI Technical Summary

Problems solved by technology

[0002] In foggy conditions, due to the scattering of atmospheric aerosol particles, the visibility of the image is severely reduced. In addition to causing a decrease in the visual perception of the image, the fog factor will also affect the accuracy of advanced tasks such as detection and segmentation. , which greatly limits the application value of the image

In addition, in the real world, it is difficult for us to obtain pairs of foggy and clear images. Due to the lack of paired foggy and clear outdoor images in the same scene as training samples, the ability of learning-based methods to remove fog is limited. Therefore, unpaired single image dehazing is very important in computer vision

[0003] Existing learning-based dehazing methods often require pairs of foggy days and clear outdoor images in the same scene as training samples, which are difficult to obtain in reality. Most of the existing non-paired dehazing methods use loop The consistency scheme enables the foggy images in the dataset to generate defogged images, and then uses the defogged images to generate foggy images, and constrains the content consistency of the foggy images before and after. At the same time, the clear images in the dataset are also similarly operate

This two-way field image generation method will increase the training cost. In addition, the cycle consistency assumes that the relationship between the foggy field and the clear field is a bijective relationship. content differences, results produced by cycle consistency may suffer from distortions in generated content

Images