From general to specific: Online updating for blind super-resolution

作者：

Highlights：

• We design the internal branch and the external branch to leverage inherent information and external priors. They could learn the degradation of the given test image and adaptively update the SR module according to learned degradation.

• An alternate optimization strategy is proposed to make super resolution reconstruction and degradation estimation more compatible.

• Extensive experiments on both synthesized and real world images show that our method can generate more visually favorable super resolution results and achieve state of the art performance on blind super resolution.

摘要

•Towards the unknown and various degradations in blind super resolution, we propose an online updating super resolution method. It could customize a specific model for each test low resolution image and thus could have more robust performance in different cases.•We design the internal branch and the external branch to leverage inherent information and external priors. They could learn the degradation of the given test image and adaptively update the SR module according to learned degradation.•An alternate optimization strategy is proposed to make super resolution reconstruction and degradation estimation more compatible.•Extensive experiments on both synthesized and real world images show that our method can generate more visually favorable super resolution results and achieve state of the art performance on blind super resolution.

论文关键词：Blind super-resolution,Online updating,Internal learning,External learning

论文评审过程：Received 31 March 2021, Revised 14 February 2022, Accepted 25 February 2022, Available online 1 March 2022, Version of Record 8 March 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108613