Research Project
This project developed a deep learning-based non-reference video quality metric incorporating psychophysical features of human visual perception. The metric supports stable prediction of subjective video quality ratings across independent datasets.
This work created a new non-reference video quality metric that incorporates psychophysical features of the user’s video experience. The metric is designed to predict subjective video quality without requiring access to an original reference video, making it suitable for real-time video monitoring and practical streaming environments.
The proposed NRspttemVQA approach integrates spatio-temporal characteristics of human visual perception into video quality prediction. Experimental results show stable performance across three independent video datasets, demonstrating consistent correlation with subjective quality ratings and improving the reliability of perceptual video quality assessment.
This project contributes to human-centred video processing, real-time media quality monitoring, and future embedded vision systems where computational efficiency, subjective quality prediction, and user experience are critical.
This project has been published as “NRspttemVQA: Real-Time Video Quality Assessment Based on the User’s Visual Perception” at the 38th International Conference on Image and Vision Computing New Zealand.
