Deep and adaptive feature extraction attention network for single image super-resolution

Abstract

Single image super-resolution (SISR) has been revolutionized by convolutional neural networks (CNN). However, existing SISR algorithms have feature extraction and adaptive adjustment limitations, leading to information duplication and unsatisfactory image reconstruction. In this paper, we propose a deep and adaptive feature extraction attention network (DAAN), which first fully extracts shallow features and then adaptively captures precise and fine-scale features by a deep feature extraction block (DFEB). It includes multi-dimensional feature extraction blocks (MFEBs) that combine large kernel and dynamic convolution layers to improve large-scale information utilization effectively. Finally, an enhanced spatial attention block (ESAB) to further selectively reinforce the transmission of details. A large number of experimental results show that our proposed model reconstruction performance is superior to existing classical methods.