Reduced-reference quality metric based on local sharpness measure for blurred images

A simple yet robust reduced-reference quality metric based on local sharpness measure for blurred images in discrete cosine transform (DCT) domain is introduced. As a single feature to represent the quantification of blur in an image, a global image sharpness is quantified from the local sharpness with a selected optimal subset of DCT coefficients in each overlapped block in the feature extraction process. Then, the final objective quality score is simply calculated by division by the reference value in the quality analysis process. The experimental results on three public quality assessment databases demonstrate the proposed method correlates highly with human subjective judgements and outperform other methods consistently in terms of prediction accuracy and monotonicity for a wide range of blurriness.