Blurriness measurement in frequency domain for image quality assessment

摘要

DCT based digital image and video compression leads to visible distortions like blockiness and blurriness, however this paper mainly focuses blurriness artifact. Subjective quality assessments are reliable but they are very costly and can't be computerized. This paper proposes three different objective quality assessment methods for blurriness estimation using full reference, reduced reference and no reference approaches. The distortion is measured in frequency domain by comparing the high frequency coefficients of the coded image. Before blurriness estimation in frequency domain, the property of Human Visual System is implemented by applying the spatial masking in spatial domain. Since the distortion is not likely to be in same amount in every part of the coded image therefore the coded image is divided into blocks and the distortion is calculated locally for each block and accumulated in the end for a single quality metric. The results show that the full and reduced reference meters are more reliable due to the availability of some reference information at receiver end. The work is tested on different set of blurred images from LIVE image database and the Pearson's correlation coefficient of 94.43% is obtained for full reference mode while it is 94.20% and 82.03% for reduced reference and no reference respectively.