Edge-aware multi-scale hybrid cascaded attention transformer for efficient image super-resolution

Abstract Image super-resolution (ISR) is fundamental for enhancing low-resolution images to high-resolution counterparts, Motivated by the need for lightweight models that deliver real-time performance without sacrificing detail preservation, this work proposes an Edge-Aware Multi-Scale Hybrid Cascaded Attention Transformer. It effectively integrates edge-focused attention mechanisms with multi-scale feature processing to advance single-image SR. By combining channel attention, multi-scale edge attention, and shifted window self-attention within a cascaded architecture, HCAT exploits complementary feature representations to better preserve fine details and sharp edges. The use of depth-wise separable convolutions further enables computational efficiency without sacrificing performance. Extensive experiments on benchmark datasets demonstrate that HCAT achieves superior quantitative and perceptual results, improving PSNR and SSIM by up to 2% over state-of-the-art methods while maintaining lightweight model complexity suitable for real-time applications.