Convolutional Neural Network with Squeeze and Excitation Modules for Image Blind Deblurring

Image blind deblurring is a classic computer vision and image processing task. It has been proved that end-to-end deep convolutional neural network (CNN) has the highest performance on nonuniform motion blind deblurring task from dynamic scene. Recently, many well designed multi-scale CNNs have been established. These CNNs process blurry images at different image resolution and improve the quality of restored image obviously. In this paper, inspired by Squeeze and Excitation Network (SENet), we introduce the core module named Squeeze and Excitation module (SE module) of SENet into our network to deblur the blurry images. We compare our network with three state-of-the-art methods on two standard datasets. Experiment results show that our proposed method outperforms other state-of-the-art methods quantitatively and visually.