Unsupervised night image defogging method using high and low frequency decomposition

Abstract

The invention provides an unsupervised night image defogging method using high and low frequency decomposition. The method comprises the following steps: decomposing an input image into a high-frequency image and a low-frequency image by using a guiding filter; combining the input image and the high-frequency image to serve as input of a fog-free image estimation network, and estimating a fog-free image; combining the input image and the low-frequency image to serve as input of a transmissivity estimation network, and estimating a transmission image; estimating an atmosphere illumination map corresponding to the input image by using a maximum filter; based on the fogless image, the transmission image and the atmospheric illumination map, reconstructing an input image by adopting an atmospheric scattering model; and taking the reconstructed loss function as a loss function, and performing end-to-end training on the network. According to the method, under the condition that paired night foggy day images and night clear images are not needed, learning and inference can be carried out only by using the observed night foggy day images, night haze can be effectively removed, and the visual performance of the night foggy day images is improved.

Description

technical field [0001] The invention relates to the technical field of computer image processing, in particular to an unsupervised nighttime image defogging method using high and low frequency decomposition. Background technique [0002] At present, many fog and haze image restoration algorithms are aimed at daytime images, such as methods based on prior information; most of the haze removal methods are based on the atmospheric scattering model, which generally assumes that the atmospheric light in the daytime is uniform, but for For night images, due to the weak ambient light and the interference of artificial light sources, the atmospheric light has changed greatly, the composition is more complex, and it is more difficult to estimate. Therefore, for night image dehazing, these algorithms will be very ineffective. [0003] The current night-time image defogging and restoration algorithms can be roughly divided into four categories: one is based on the experience of daytime...

AI Technical Summary

Problems solved by technology

[0004] However, the current night-time image dehazing methods based on deep learning are trained in a supervised manner, which requires a large number of pairs of night-time clear images and foggy image data sets, and the performance of the dehazing model obtained by this method It is directly related to the quality of the data set; in addition, because the dehazing method based on deep learning requires a large amount of data, it takes a lot of time to train the dehazing model

Images