Border Compensation Scheme for Image Deblurring
Jian–Jiun Ding, Shiang–Chih Hua · 2021 IEEE 3rd Eurasia Conference on IOT, Communication and Engineering (ECICE) · 2021
Border compensation is critical for many image processing problems, including image deblurring. Without a proper border compensation mechanism, the border is misidentified as a high-frequency component, and artifacts may be generated after image deblurring. There are several existing border compensation methods, including zero padding, border repetition, mirror reflection, and slope extension. However, all of these methods inevitably have artifact problems due to the discontinuity of higher-order differences. In this work, we propose an alternative border compensation model by using the Gaussian extension model. Since the derivative of the Gaussian function at the center is zero no matter what the order of the derivative is, using the proposed algorithm much reduces the artifact around the border. Experiment results show that, with the proposed border extension scheme, the image deblurring results have much less artifact. It is helpful for image quality improvement and computer vision.