Dynamic scene blind image deblurring based on local and non-local features

Blind image deblurring is a fundamental and challenging task in the field of computer vision. Despite image deblurring has been made considerable progress, there is still room for improvement in the visual effect and details of the images. Therefore, we present an image deblurring model based on local and non-local features for non-uniform scene deblurring in an end-to-end fashion. Correspondingly, we develop a dense dilated block (DDB) and an improved attention module (IAM) to excavate local and non-local features, respectively. DDB focuses on enhancing feature correlation and constructing complex features in high dimensions by exploiting local features. IAM is a gate mechanism, which implicates spatial context information and attention maps based on non-local channels dependencies. Compared to the previous methods, our method surpasses state-of-the-art (SOTA) methods on both synthetic datasets and real-world images.