Layer Priors and Encoding-decoding Network for Image Dehazing
Chu Zhao, Hang Li, Man Jiang · DOAJ (DOAJ: Directory of Open Access Journals) · 2025
In order to solve the shortcomings of the current image dehazing algorithm, which has poor recovery effect and general timeliness, a novel image dehazing algorithm combining the layer priors and encoding-decoding network is proposed. Firstly, the haze image is divided into background layer and haze layer, and the time gradient of background layer and horizontal gradient of haze layer and the pre-trained Gaussian mixture model corresponding to each layer are used as the prior conditions to construct the model function. Then, a channel attention module is added at the end of the encoder and the beginning of the decoder to assign different weights to the haze related feature maps extracted by the encoder and calculate the transmittance accurately. Thirdly, using the proposed fuzzy partition entropy graph cutting algorithm, the transmittance is divided into close-range, mid-range and far-range under different scene light coverage. The experimental results show that the new algorithm has a good dehazing effect on both synthetic and real fog maps compared with other dehazing methods.