Dual-lens Reference Image Super-Resolution

Jing Zhu (Samsung Research America), Wenbo Li (Samsung Research America),* Hongxia Jin (Samsung Research America)
The 33rd British Machine Vision Conference


Reference based image super-resolution (RefSR) methods have been demonstrated to be effective, and these methods use a single high-resolution (HR) image as the reference. Nowadays, most high-end mobile devices are embedded with three lenses of different fields of view. The broader field of view a lens captures, the lower image resolution it yields. In this fashion, the wide-angle lens captures the broadest field of view but yields the lowest image resolution. In order to obtain the epic wide-angle photo for large-screens, e.g., TV, this paper studies how to improve the resolution of the wide-angle lens using the photos captured by the other two lens as references, i.e., dual-lens reference. To start with, we propose a base model which treats the dual-lens references as the separate auxiliary inputs, and observes obvious performance gain compared to the existing RefSR methods based on the single reference. Then, we propose an enhanced model which exploits the relationships among the dual-lens references and the source wide-angle image, and we found that the mining and usage of their relationships are beneficial. We conduct the experiments on a real dataset captured by multi-lens in a phone and six additional datasets which are simulated based on publicly available datasets.



author    = {Jing Zhu and Wenbo Li and Hongxia Jin},
title     = {Dual-lens Reference Image Super-Resolution},
booktitle = {33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
publisher = {{BMVA} Press},
year      = {2022},
url       = {https://bmvc2022.mpi-inf.mpg.de/0359.pdf}

Copyright © 2022 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection