End-to-end network image defogging method based on Retinex theory

Abstract

The invention discloses an image defogging method of an end-to-end network based on a Retinex theory. The method mainly comprises the following steps: firstly, establishing a defogging network, then embedding a Retinex model into the defogging network to realize end-to-end learning, extracting feature information from a foggy image to estimate a brightness image and a reflection image, and furtherrecovering a clear image according to the Retinex model. By using the network in the invention, the brightness image and the reflection image of the image can be jointly estimated, and then the fog-free image is recovered according to the Retinex model. The defogging network provided by the invention does not depend on an atmospheric scattering model; therefore, the problem of image quality reduction after defogging caused by inaccurate transmission image estimation is avoided; and the Retinex theory is a color perception model based on human vision, and the network recovers the fog-free image by estimating the brightness map and the reflection map, so that the fog-free image is more consistent with the visual law of human eyes, and the recovered fog-free image is clearer and more natural.

Description

technical field [0001] The invention belongs to a computer image processing method, in particular to an image defogging method. Background technique [0002] Outdoor images taken under fog, haze and other weather conditions, due to interference such as absorption and scattering by atmospheric suspended particles, will produce degradation phenomena such as contrast drop, blurred details, color distortion, etc., which will seriously affect visual systems such as outdoor video surveillance and target recognition. performance. Therefore, image dehazing is particularly important in the field of computer vision applications and digital image processing. [0003] With the wide application of convolutional neural network (CNN) in computer vision tasks, methods based on deep learning have become the mainstream direction of image dehazing research. This method directly learns the mapping relationship between the foggy image and the non-foggy image to achieve dehazing, or estimates p...

AI Technical Summary

Problems solved by technology

[0004] Although these methods have obtained better dehazing results, the method [1-2] Directly learning the mapping relationship between the foggy image and the fog-free image to restore the fog-free image lacks a theoretical basis, and there is a phenomenon of image distortion after defogging

Although the method [3-5] The haze-free image is restored according to the atmospheric scattering model, but the transmission map estimated by the network often contains too much detail information, and most methods set the atmospheric light value as a globally consistent constant, resulting in the degradation of the restored haze-free image

Images