Image Super-Resolution and Deblurring Using Generative Adversarial Network
Bianli Du, Xiaokang Ren, Saijian Chen, Jie Ren, Danling Cao · 2019
Image super-resolution and deblurring are two highly ill-posed problems that are usually dealt separately. However, real-world images are often low-resolution and have complex blurring. This paper focuses on ordinary natural scene images and reconstructs clear high-resolution images directly from blurred low-resolution inputs. Firstly, we propose a model based on generative adversarial network to jointly process image super-resolution and non-uniform motion deblurring. Secondly, we decouple this joint problem into feature extraction module, super-resolution reconstruction module and deblurring module. The modules promote each other and reconstruct clearer high-resolution images. Finally, we use bilinear interpolation followed by a convolutional layer to achieve upsampling instead of using the common deconvolution layer, which effectively suppresses checkerboard artifacts. The experimental results show that the proposed method is efficient and can perform better than the existing advanced algorithms in both quantitative and qualitative performance.