A Wavelet-based Progressive Super-Resolution Model

Deep-learning based Super-Resolution (SR) methods have achieved remarkable performance in many aspects. However, most existing methods are constructed for a single specific scale, and few are multi-scale, lightweight, and have a short running time. To this end, we propose a wavelet-based progressive super-resolution method (WPSR). WPSR could process low and high-frequency information separately by reconstructing different wavelet coefficients in order and could generate multiple SR images at multiple scales sequentially. Experiments demonstrate that our WPSR could outperform existing state-of-the-art methods at multiple scales, especially for ×8 and ×16 scale factors.