The goal of image sharpening is to enhance details that are blurred. Fine details in the frequency domain correspond to high frequencies. Therefore image sharpening can be achieved by the use of high-pass filters. Image sharpening was performed using Ifranview. Ifranview uses its own sharpening value scale, which ranges from 0 to 100. We compiled data for sharpening values from 0 to 80.
Both algorithms give a good correlation on the extracted watermark throughout the image sharpening attack. For this image Cox is a little bit stronger than the Xie algorithm. Xie's correlation has a steeper slope than that of Cox. From the images above we see that at a sharpening value of 80 the image has been degraded by a considerable amount visually, but the watermark for both algorithms can still be recovered with a high correlation.
The purpose of this next test is to simulate scanning the watermarked image. Therefore the image is rotated by a small angle and cropped. The rotation and cropping are done to simulate the misalignment when scanning an image. The image was rotated both clockwise and anti-clockwise to compare the difference if any. The images were generated using Stirmark.
Figure #16: Xie Image Scanning Simulation
Figure #18: 0.25 and 2 Degree Rotated and Cropped Images
From the above results we see that the Cox algorithm is more resilient to this type of attack than Xie. The rotation and cropping combination on the image affects the reconstruction point of the Xie watermark extraction method leading to an unsatisfactory correlation. The Cox algorithm has a good correlation on the extracted watermark up until a rotation of 2 degrees with cropping.
Images used for this test were generated with Stirmark. The image is rotated followed by cropping and then resizing to keep original scale. The rotated angles are larger than the ones used before and are as follows: 5, 10, 15, 30, 45, and 90.
The combination of rotation and cropping attack destroys the watermark recover process completely which is expected from results obtained before. Unsatisfactory results were expected because rotation or cropping alone degrades the watermark considerably for reasons given under those testing sections.