The High-Efficiency Motion-Blurred Video Restoration Method by Feature Enhancement-Based Deep Learning Network
Chao-Ho Chen, Chia-En Lin, Tsong-Yi Chen, Da-Jinn Wang, Hsueh-Liang Hsiao, Cheng-Kang Wen · 2022 IEEE 11th Global Conference on Consumer Electronics (GCCE) · 2022
To achieve high-efficiency motion-blurred video restoration for hand-held cameras, this paper presents a feature enhancement-based deep learning network which mainly employs the channel attention layer to generate the most effective features for improving the image-deblurring effect. In addition, by weighting the color saturation of the input image of training process to preserve its original clarity, the inherent problem of turning the clear part of an input image into the fuzzy part, caused by the inappropriate convolution results in the previous video deblurring methods using deep learning networks, can be substantially reduced. Thus, if more feature information can be effectively extracted, the feature extraction process does not require much computational effort. Experimental results show that the proposed method has lower computational complexity and higher restoration quality, especially for the numbers and characters, compared to other deep learning network-based methods.