Emotion and saliency in visual quality assessment

Dong, Zhengyan 2025. Emotion and saliency in visual quality assessment. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 25 November 2026 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (16MB) | Request a copy
	PDF - Supplemental Material Restricted to Repository staff only Download (251kB) | Request a copy

Abstract

This research investigates human-centric Visual Quality Assessment (ViQA) by integrating emotional perception, saliency, and temporal dynamics into image and video quality modelling. In the first part, we examine how emotional content affects subjective judgements of image quality. To support this, we introduce the Cardiff University Emotional Image Quality Assessment (IQA) Database (CUEAD), a benchmark of 600 images distorted by motion blur, JPEG compression, and jitter at three intensity levels. The images, selected from Emotion6, are rated by expert observers to ensure perceptual reliability. Based on this dataset, we propose a deep no-reference IQA model featuring a fusion mechanism that integrates emotion-related style features with traditional quality indicators. Two pretraining datasets provide prior knowledge for emotion-aware and distortion-aware components. Experiments demonstrate the model’s advantage in predicting perceptual quality across varied emotional contexts. While this part addresses static images, the broader challenge lies in modelling how human attention and quality perception change over time, particularly in videos. Through the study of video saliency, we further examine how spatial attention, motion, and quality interact temporally, identifying key limitations in current evaluations that average frame-level performance and neglect temporal variation. We highlight the need for content-aware, temporally sensitive evaluation protocols. To address these issues, we propose the Frame-Quality-Aware Dynamic Video Saliency (FQA-DVS) model, which integrates spatial, temporal, and quality-driven cues via a dual-branch architecture. FQA-DVS improves saliency prediction in low-quality and dynamic video scenarios. By jointly modelling perceptual factors such as emotional influence, attention shifts, and content distortion, this study offers a unified framework for visual quality assessment, advancing both IQA and Video Quality Assessment (VQA) toward more robust, context-aware, and perceptually aligned models suitable for real-world multimedia applications.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software
Date of First Compliant Deposit:	25 November 2025
Date of Acceptance:	17 November 2025
Last Modified:	08 Dec 2025 15:42
URI:	https://orca.cardiff.ac.uk/id/eprint/182630

Actions (repository staff only)

Edit Item