From Distortion Manifold to Perceptual Quality: a Data Efficient Blind ImageÂ Quality Assessment Approach

作者：

Highlights：

• We explore to improve generalize ability of blind IQA models with limited training data

• We propose to explicitly learn a distortion manifold

• A novel model proposed to extract both low level distortion patterns and high level semantic information

• An effective training framework including a masked labeling strategy and a gradual weighting curriculum

摘要

•We explore to improve generalize ability of blind IQA models with limited training data•We propose to explicitly learn a distortion manifold•A novel model proposed to extract both low level distortion patterns and high level semantic information•An effective training framework including a masked labeling strategy and a gradual weighting curriculum

论文关键词：Image quality assessment,No-Reference,Generalizability,Distortion manifold

论文评审过程：Received 19 March 2022, Revised 2 September 2022, Accepted 15 September 2022, Available online 17 September 2022, Version of Record 22 September 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.109047