This paper proposes Frequency-Guided Attention (FGA), a lightweight upsampling module for single-image super-resolution. Conventional upsamplers, such as subpixel convolution, while efficient, often fail to reconstruct high-frequency details and introduce aliasing artifacts. FGA addresses these issues by incorporating (1) a Fourier feature-based multilayer perceptron (MLP) for positional frequency encoding, (2) a cross-resolution correlated attention layer for adaptive spatial alignment, and (3) a frequency-domain L1 loss for spectral fidelity supervision. With only 0.3M additional parameters, FGA consistently improves performance across five different super-resolution backbones in both lightweight and full-capacity scenarios. Experimental results demonstrate average PSNR gains of 0.12–0.14 dB and up to 29% improvement in frequency-domain coherence, particularly on texture-rich datasets. Visual and spectral evaluations confirm that FGA is effective in reducing aliasing and preserving fine details, demonstrating that it is a practical and scalable alternative to existing upsampling methods.
