Multi-Scale Feature Fusion Network with Attention for Single Image Dehazing

In this paper, we propose an end-to-end trainable multi-scale feature fusion network with attention (MSFFA-Net) to directly restore the clean image from single hazy image. The proposed dehazing method does not rely on the atmosphere scattering model. Firstly, the backbone of the proposed MSFFA-Net i...

Full description

Saved in:

Bibliographic Details
Published in:	Pattern recognition and image analysis 2021-10, Vol.31 (4), p.608-615
Main Author:	Hu, Bin
Format:	Article
Language:	eng
Subjects:	Analysis Computer Science Image Processing and Computer Vision Image restoration Mathematical Theory of Images and Signals Representing Multiscale analysis Pattern Recognition Pixels Processing Recognition Understanding
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	In this paper, we propose an end-to-end trainable multi-scale feature fusion network with attention (MSFFA-Net) to directly restore the clean image from single hazy image. The proposed dehazing method does not rely on the atmosphere scattering model. Firstly, the backbone of the proposed MSFFA-Net is a multi-scale grid network (GridNet) that allows efficient information exchange across different scales, and can effectively alleviate the bottle-neck issue often encountered in the conventional multi-scale approach. A channel-wise attention mechanism is used to fuse the feature from row stream and column stream to flexibly adjust the contributions from different scales in feature fusion. Secondly, a feature fusion structure which combines the channel fusion and pixel fusion in channel-wise and pixel-wise features is used to learn more weight from important features, and the structure is a basic module of GridNet. Experimental results in RESIDE dataset indicate that the proposed method outperforms the state-of-the-art both quantitatively and qualitatively.
ISSN:	1054-6618 1555-6212