Quality Assessment of Sharpened Images

This master’s thesis investigates the possibilities of quality assessment of sharpened images. Reliable assessment of these images is crucial for comparing the performance of particular sharpening techniques and also for finding the optimal amount of sharpening. In the thesis, fundamentals of edge detection, sharpness perception, and sharpening algorithms are given. Four of the algorithms are described in more detailed view and subsequently employed in creation of testing images. Furthermore, the pair comparison method for subjective quality tests and procedures for increasing its applicability are discussed. An overview of existing objective quality criteria, focusing at the metrics suitable for sharpened images, and several augmentations of one of the metrics are also presented. Eventually, a new way of incorporating full-reference metrics into assessment of sharpened images, using intentional over-sharpening, is considered. In the experimental part of the thesis, two pre-tests and an extensive subjective quality test are described. Adaptive pair comparison method was used for this test. The results of the test are then the ground-truth of two Receiver Operating Characteristics (ROC) Analyses. 19 state-of-the-art metrics and 7 proposed augmentations are analyzed. From the outcome of the analyses, it can be concluded that the proposed enhancements lead to an improvement of metric’s performance which thereafter outperforms the rest of the state-of-the-art metrics.