The perceptual quality of digital imagery is of great interest in many applications. Blur artifacts can be among the most
annoying in processed images and video sequences. In many applications of perceptual quality assessment, a reference is
not available. Therefore no-reference blurriness measures are of interest. In this paper, we present a universal, reference-free
blurriness measurement approach. While some other methods are designed for a particular source of blurriness such
as block-based compression, the proposed is universal in that it should work for any source of blur. The proposed
approach models the gradient image of the given image as Markov chain and utilizes transition probabilities to compute
a blurriness measure. This is the first time that transition probabilities are applied to perceptual quality assessment.
Specifically, we first compute the transition probabilities for selected pairs of gradient values and then combine these
probabilities, using a pooling strategy, to formulate the blurriness measure. Experimental studies compare the proposed
method to the state-of-the-art reference-free blurriness measurement algorithms and show that the proposed method
outperforms the commonly used measures.