A comparison of image sharpness metrics and real-time sharpening methods with GPU implementations

Abstract

Improving the quality of an image increases the probability, speed and accuracy with which possible objects of interest can be located and identified. Image sharpening, in particular, can correct for soft focus and strengthen the outlines of objects thus improving the identification and segmentation both by automatic means and by man in the loop systems. Output pixel independence is ensured so that a GPU can be used to sharpen the pixels in parallel, achieving processing performance increases of 20--360 fold. This work provides a metric which can quantify the sharpness of an image and shows that the sharpness of live video can easily be doubled in realtime on commercial desktop computers without inducing excessive noise.