Multi-Scale Feature Fusion and Structure-Preserving Network for Face Super-Resolution

Deep convolutional neural networks have demonstrated significant performance improvements in face super-resolution tasks. However, many deep learning-based approaches tend to overlook the inherent structural information and feature correlation across different scales images, making accurate recovery of structure low-resolution cases challenging. To address this, this paper proposes a method that fuses multi-scale features while preserving facial structure. It introduces novel residual block (MSRB) reconstruct key parts structures from spatial channel dimensions, utilizes pyramid attention (PA) exploit non-local self-similarity, improving details reconstructed face. Feature Enhancement Modules (FEM) are employed upscale stage refine enhance current using previous stages. The experimental results on CelebA, Helen LFW datasets provide evidence our achieves superior quantitative metrics compared baseline, Peak Signal-to-Noise Ratio (PSNR) outperforms baseline by 0.282 dB, 0.343 0.336 dB. Furthermore, demonstrates improved visual two additional no-reference datasets, Widerface Webface.