Open AccessDynamic Scene Deblurring of Multi-Scale Progressive Attention Network
Author(s) -
Wenzhuo Huang,
Yueming Deng,
Yiming Liu,
Jun Wu,
Xiaojun Li
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1880/1/012007
Subject(s) - deblurring , computer science , artificial intelligence , feature (linguistics) , pyramid (geometry) , computer vision , enhanced data rates for gsm evolution , residual , pattern recognition (psychology) , scale (ratio) , image (mathematics) , image restoration , image processing , algorithm , mathematics , linguistics , philosophy , physics , geometry , quantum mechanics
In order to enhance the attention to the image foreground targets in dynamic scenes and better suppress the generation of image edge artifacts, a multi-scale progressive attention network (MSPA) deblurring algorithm is proposed, MSPA is based on the GAN structure. In the feature extraction stage, MSPA designs multi-scale width perception and pyramid perception residual blocks to help our skeleton network better extract multi-scale local features and reduce the difficulty of network training, and In order to better obtain the semantic information of the image, the channel attention mechanism is used to correct the deep features of the blurred image; in the feature fusion stage, the progressive channel and spatial attention network (PCSA) is designed to enhance the receptive field, and selectively integrate multiple levels semantic information and enhance the non-local connection between features in progressive method. A large number of experiments show that the MSPA algorithm is superior to many advanced algorithms on the GoPro and Kohler datasets.