Enhancing lightweight image super-resolution through frequency-enhanced mamba networks

Image super-resolution (SR) aims to enhance low-resolution images to high-resolution counterparts with superior quality. Despite notable advancements using deep learning, maintaining computational efficiency while capturing long-range dependencies remains challenging. Inspired by state-space models, we introduce the frequency enhanced Mamba network (FEMN), designed to integrate global and local image features across frequency and spatial domains. Our approach incorporates a frequency feature extraction module (FFEM) for adaptive frequency cue extraction and a localized spatial aggregation module (LSAM) for enhanced spatial variation representation. By synergistically combining FFEM and LSAM, FEMN effectively bridges frequency-aware global modeling with fine-grained local spatial details, enabling more comprehensive feature learning. Extensive experiments on standard benchmarks demonstrate FEMN’s superior performance and efficiency, achieving state-of-the-art reconstruction quality while maintaining low computational complexity. The code is available at https://github.com/Qiuyu2024/FEMamba/tree/main.