In recent years, multi-scale approach has been applied to image restoration tasks, including super-resolution, deblurring,etc., and has been proved beneficial to both optimization-based methods and learning-based methods to improve therestoration performance. Meanwhile, it is observed that high-frequency information plays an important role in blindmotion deblurring. Unlike previous learning-based methods, which simply deepen deblurring network withoutdiscriminating the low-frequency contents and the high-frequency details, we propose a novel multi-scale convolutionalneural network (CNN) framework with residual channel attention block (RCAB) for blind motion deblurring. RCAB hasthe residual in residual (RIR) structure, which consists of several residual groups with long skip connections and allowslow-frequency information pass through the skip connections conveniently, and can adaptively learn more usefulchannel-wise features and pay more attention to high-frequency information. Experimental results show that ourproposed method can obtain better deblurring images than state-of-the-art learning-based image deblurring methods interms of both quantitative metrics and visual quality.
展开▼
