A simple spatial domain method for quality evaluation of blurred images

In this paper, a simple yet highly accurate algorithm for No-Reference quality estimation of blurred images is proposed. The proposed work is motivated by the fact that when blurring occurs, regions with large pixel variations are likely to be affected more than the regions with small pixel variations. Since the human visual system is also more attentive to distortions in the regions with large pixel variations; therefore, it will be advantageous to exploit them for blurriness estimation. Moreover, blurring also causes loss of details, thereby resulting in the decrease in mean pixel variation and the maximum pixel variation. It is also observed that the ratio of mean to maximum pixel variation increases with blurriness. Motivated by these facts, the maximum pixel variation and mean pixel variation are utilized to estimate the quality of the image affected by the blurriness. The proposed algorithm is highly competitive and outperforms most of the state-of-the-art algorithms both in time-complexity as well as accuracy over various standard databases.