Abstract Dual-lens Super-Resolution (Dual-lens SR), as a special case of Reference-based Super-Resolution (RefSR), aims to reconstruct high-frequency details for wide-angle images aided by telephoto images. In this paper, we address two challenging issues in Dual-lens SR: (1) how to perform more accurate feature matching between wide-angle and telephoto images; (2) how to transfer reference textures and fuse them with low-resolution images effectively. To resolve these challenges, we propose a semantic-texture fusion-based feature matching (STF-matching) module that significantly improves the accuracy of feature matching. Furthermore, we develop a reference feature transfer and aggregation block (RFTAB) to adaptively transfer reference textures and reconstruct high-frequency details. Extensive experiments on public benchmarks demonstrate that our method achieves superior performance over current state-of-the-art (SOTA) methods in both quantitative and qualitative evaluations.
Kang et al. (Mon,) studied this question.