CMOA-Net: Competent Multi-Observant Attention Network for Single-Image Super-Resolution

Recent advancements in deep convolutional neural networks show significant improvements in single-image super-resolution (SR). Existing SR methods typically focus on designing deeper or wider network architectures to enhance performance and fail to effectively utilize the low-resolution image features. However, these approaches often suffer from high computational costs. Additionally, several CNN-based methods face challenges in capturing adequate spatial context. We proposed the lightweight Competent Multi-Observant Attention Network (CMOA-Net) to address these issues and achieve more robust multi-scale features and feature correlations. The CMOA-Net includes an Efficient Feature Extraction Network (EFEN) that incorporates the Globalized Multi-Discerning (GMD) block and Competent Spatial Attentiveness (CSA). The proposed model is designed to enhance the representation capabilities of multi-scale features. Furthermore, The CSA module uses spatial and channel attention to capture long-range dependencies and preserve key features, enhancing output image quality. The effect of these combined components significantly enhances the proposed network performance, resulting in exceptional accuracy. Results reveal that the proposed state-of-the-art outperforms five synthetic benchmark datasets.