[2603.21045] LPNSR: Prior-Enhanced Diffusion Image Super-Resolution via LR-Guided Noise Prediction

Computer Science > Computer Vision and Pattern Recognition arXiv:2603.21045 (cs) [Submitted on 22 Mar 2026] Title:LPNSR: Prior-Enhanced Diffusion Image Super-Resolution via LR-Guided Noise Prediction Authors:Shuwei Huang, Shizhuo Liu, Zijun Wei View a PDF of the paper titled LPNSR: Prior-Enhanced Diffusion Image Super-Resolution via LR-Guided Noise Prediction, by Shuwei Huang and 2 other authors View PDF HTML (experimental) Abstract:Diffusion-based image super-resolution (SR), which aims to reconstruct high-resolution (HR) images from corresponding low-resolution (LR) observations, faces a fundamental trade-off between inference efficiency and reconstruction quality. The state-of-the-art residual-shifting diffusion framework achieves efficient 4-step inference, yet suffers from severe performance degradation in compact sampling trajectories. This is mainly attributed to two core limitations: the inherent suboptimality of unconstrained random Gaussian noise in intermediate steps, which leads to error accumulation and insufficient LR prior guidance, and the initialization bias caused by naive bicubic upsampling. In this paper, we propose LPNSR, a prior-enhanced efficient diffusion framework to address these issues. We first mathematically derive the closed-form analytical solution of the optimal intermediate noise for the residual-shifting diffusion paradigm, and accordingly design an LR-guided multi-input-aware noise predictor to replace random Gaussian noise, embedding LR ...