IOPNet: Profiling individual opinions in image quality assessment

Ma, Yueran, Wang, Huasheng, Liu, Jiang, Corcoran, Padraig ORCID: https://orcid.org/0000-0001-9731-3385, Whitaker, Roger M. ORCID: https://orcid.org/0000-0002-8473-1913, Allen, Stuart M. ORCID: https://orcid.org/0000-0003-1776-7489, Colombo, Gualtiero B., Zhou, Wei, Zhai, Guangtao, Chen, Ying and Liu, Hantao ORCID: https://orcid.org/0000-0003-4544-3481 2026. IOPNet: Profiling individual opinions in image quality assessment. IEEE Transactions on Multimedia 10.1109/tmm.2026.3739488

Full text not available from this repository.

Abstract

Image quality assessment (IQA) aims to quantitatively measure human perception of visual quality. Existing IQA methods focus on the mean opinion score (MOS), which does not capture the variability in human perception shaped by personal preference from individual human differences. To predict such perceptual distribution, in this paper, we introduce the Individual Opinion Profiling (IOP) framework, a novel approach that measures individual preferences and biases in assessing image quality. By measuring the full spectrum of individual ratings, IOP provides a more comprehensive quantification of diverse viewer experiences of image quality that can benefit downstream tasks. To achieve this, we propose a deep learning-based IOP network (IOPNet) that employs the Swin Transformer and Graph Attention Blocks to extract robust representations from images. We utilise a patch hierarchical perception regression framework to predict individual IQA ratings, with an output dimension that flexibly adapts to accommodate the required number of viewers. Additionally, we propose two novel metrics to evaluate the accuracy of the prediction of individual opinions to facilitate model comparison. Experimental results validate the effectiveness of our proposed method and the significant benefits that it yields for image quality assessment when incorporating the profile of individual opinions.

Item Type:	Article
Date Type:	Published Online
Status:	In Press
Schools:	Schools > Computational & Mathematical Sciences Schools > Computer Science & Informatics
Publisher:	Institute of Electrical and Electronics Engineers
ISSN:	1520-9210
Last Modified:	12 Oct 2026 13:45
URI:	https://orca.cardiff.ac.uk/id/eprint/190161

Actions (repository staff only)

Edit Item