Enhanced Super-Resolution Using Cross Attention: Refining HMA for Local Image Restoration

In this paper, we propose a novel method that integrates Cross Attention into the existing Hybrid Multi-Axis Aggregation Network for Image Super-Resolution(HMANet) to improve local super-resolution accuracy. While previous HMANet methods primarily focused on enhancing the resolution of the entire image, our approach emphasizes local image regions for more detailed restoration. By leveraging Cross Attention for context interaction, we achieve localized super-resolution with a focus on specific parts of the image. Our experiments demonstrate that the proposed method outperforms existing approaches in terms of accuracy, and shows promising results when evaluated with different loss functions. Our source code is available at https://github.com/msdsmlhmaca-for-Iocal-image-restoration.