Existing and emerging image quality metrics

This paper summarizes and evaluates some of the existing methods of measuring and quantifying the quality of a digital image. Unfortunately, no general method has been found. The performance of a quality metric is normally gauged by its prediction accuracy, monotonicity and consistency. It is also expected to mirror the quality scores assigned by independent human observers. Research to this point has generally focused on full-reference (FR) measures that assume that coded and original images are available. Often times, an original is not easily obtainable, or perhaps does not even exist. Therefore, researchers have recently shown a great deal of interest in developing reduced-reference (RR) and no-reference (NR) metrics. This study implements and compares some of the most common IQMs and seeks to determine if there is any difference in their performance. Analysis of the results focuses on determining if any IQM is superior to the others over a general set of test images