| Summary: | 碩士 === 亞洲大學 === 資訊傳播學系碩士班 === 99 === Digital technologies have been progressively developed today, while the digital image communication is also widely used. How to improve the performance of an image denoising system is an important research task. Most image denoising algorithms cannot well recover the heavy noise corrupted image. In this study, we attempt to propose a novel image denoising algorithm which can efficiently remove the infected noise, even at the heavy noise corruption (with noise density 80%). Initially, we classify a central pixel in a window to either noise dominated (pixel value is 0 or 255) or signal dominated. This is performed by analyzing the motion direction and the motion variation of the central pixel (12 candidate directions). If the motion variation on the optimum direction is below a threshold, the central pixel is classified as a clean image pixel. It is kept unchanged to maintain image quality. Conversely, if the motion variation on the optimum direction exceeds a threshold, the central pixel is classified as a noisy pixel. This pixel is replaced by the weighted median value on the optimum direction, thus the noise is removed. In addition, the directional median filter is iteratively performed to remove greater amount of noise. Experimental results show that the proposed optimum-directional-switch-median filter cannot only efficiently suppress high-density impulse noise, but also can reserve the detailed information of an image. It results in the denoised image being clear and free from blur effect.
|
|---|
