No-Reference Image Sharpness Metric Based On Directional Derivatives

In this paper, we present a no-reference image sharpness metric using the directional derivatives. The main idea is from the meaning of sharp which involves a sudden or abrupt change in direction or course. In other words, people have a feeling of sharpness when they receive sudden changes in some direction during their observations of an image. The metric is composed of three steps. First, the directional derivatives of the input image are computed by directional derivative filters. Then the maximum and minimum of these directional derivatives are selected to construct the absolute changes and relative changes which are proportional to the sudden changes received by an observer. Finally, the sharpness metric is determined by a weighted average of these changes. Our experiment on three simulated blur images demonstrates the proposed sharpness metric is competitive the relevant sharpness metrics for evaluating both real and synthetic blurring.