Benchmarking study demonstrates superior prediction of subjective image quality across benchmark databases, indicating improved alignment with the human visual system.
Key Points
To develop a full-reference image quality assessment framework that closely mirrors the human visual system by amplifying subtle degradations and accounting for geometric discrepancies.
Transformed reference and distorted images into wavelet subbands and color histograms to emulate human visual perception.
Applied collaborative feature refinement to emphasize subtle visual degradations and calculated Hausdorff distance to evaluate local geometric distortions.
Evaluated metric performance against subjective quality ratings across multiple standard image benchmark databases.
Accurately predicted subjective human quality ratings under both subtle distortions and severe visual artifacts.
Outperformed existing state-of-the-art full-reference quality metrics in alignment with human visual evaluation across multiple benchmarks.