Study of Saliency in Objective Video Quality Assessment

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Signal Processing

Background:

• Predicting human-perceived video quality is challenging and crucial for practical applications.
• Visual saliency is a key factor in video quality assessment (VQA), but reliable data acquisition and metric integration remain issues.
• Current objective video quality metrics (VQMs) often lack robust incorporation of visual saliency.

Purpose of the Study:

• To develop a refined methodology for reliably collecting eye-tracking data for video quality research.
• To investigate the impact of visual saliency on the performance of objective video quality metrics (VQMs).
• To propose an optimal approach for integrating saliency into VQMs and compare computational saliency models with ground truth data.

Main Methods:

• A large-scale eye-tracking experiment with 160 observers and 160 video stimuli was conducted.
• A novel data collection methodology was employed to eliminate bias in eye-tracking measurements.
• Visual saliency data was integrated into established VQMs, and their performance was evaluated.

Main Results:

• The proposed methodology yielded reliable eye-tracking data for saliency measurement.
• Integration of saliency significantly improved the performance of several VQMs.
• A novel approach for optimal saliency utilization in VQMs was devised and validated.
• The study assessed the effectiveness of computational saliency models versus ground truth data.

Conclusions:

• Reliable eye-tracking data collection is feasible with the proposed methodology.
• Visual saliency is a critical component for enhancing objective video quality metrics.
• The findings provide a foundation for developing more accurate and human-aligned VQMs.