.png){kind=icon}
.png){kind=icon}
Return
Progressive encoding-decoding image dehazing network
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-15638-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In this paper, we propose a progressive encoding-decoding network (PEDN) for image dehazing. First, we built a basic dehaze unit to progressively process the image to achieve image dehazing in stages. The basic dehaze unit is composed of a feature memory module and an encoding-decoding network. The feature memory module is used to transfer features at different progressive stages. The encoding-decoding network is responsible for feature extraction, encodes and decodes images by fusing different levels of pyramid features. The basic dehaze unit shares parameters during the progressive process, which effectively reduces the difficulty of network training and improves the fitting speed. The proposed model is an end-to-end image dehazing network, which does not depend on the atmospheric scattering model. In addition, we extracted the depth information of the hazy image and obtained its pyramid features, and incorporated the depth information into the feature extraction to guide the network to restore clear images more accurately. Experiments show that the our method not only performs well on synthetic datasets, but also has excellent performance on real world hazy images. It is superior to current image dehaze methods in quantitative indexes and visual perception. Code has been made available at https://github.com/LWQDU/PEDN.
Keywords:
Image dehazing
Deep convolutional network
Scene depth
Real scene
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W