Automated image quality assessment methods can be
roughly categorized as three groups: full-reference, reducedreference
and no-reference methods [1]. The full-reference
method assumes a high quality (un-degraded) image is available
and the degraded image can be compared with its corresponding
high quality image. The classic full-reference
metrics are peak Signal-to-Noise ratio (PSNR), mean squared
error (MSE), Signal-to-Noise ratio (SNR), and Pearson correlation
[2]. Reduced-reference methods use features instead
of full images to compare the degraded image and its corresponding
high quality image such as the one proposed in
[3]. Both full- and reduced-reference assessment techniques
require high quality reference images that are not available
in most applications. Therefore, the no-reference quality
assessment which measures the quality of the images based
on their properties alone is a critical requirement for such
applications.