Study On Image Denoising Based On Cloud Model And Its Application

How to remove impulse noise has been studied as a classic problem, because the noise often corrupts the image pixels. At the beginning, the traditional filters unconditionally fulfill on each pixel without considering whether the pixel is corrupted or not. As the uncorrupted pixels are altered, image details were damaged and image quality was sharply reduced specially in high noise levels. Then, people present the switching filter as novel solution to remedy this. They try to identify the noise pixels before filtering, thus only the noise pixels have been replaced. This makes a great improvement in image denoising.Uncertainties are inherent features. Among the uncertainties involved in impulse noise, the randomness and the fuzziness are the two most important features. The traditional filters only take the randomness into account, but the switching filters focus on them both. However, unfortunately, for the switching filters, the membership degrees are too "hard", which makes the filter creates many mistakes in noise detection. In addition, the switching filters have some blindness to noise detection. They only consider the extreme values, the median value, and the detected pixels in noise detection, yet ignore the others. So, they fail to tell the difference between the noise and the uncorrupted pixels. This reduces the accuracy of the noise detection.In image restoration, most switching filters do not consider the following issues. Just like, for the pixel, different gray values or locations have different contributions to image restoration; anisotropy effects in image restoration, etc. These make the filters cannot restore the image with good detail-preservation.Aiming to solve these problems, this paper gives some relevant solutions, and the main contents can be divided into three parts.Firstly, through detailed analysis of MMEM filter, we find the deficiency of the filter (Just like, the window size is too small, the global threshold is too "hard", and the AVG value is wrong, etc.). Thus the paper provides some relevant solutions (such as, an adaptive window, a "soft" local threshold, and a novel value range for the AVG, etc.). Based on this, the paper presents a novel filter, adaptive MMEM filter. The experimental results show that the adaptive MMEM filter makes a great improvement in noise detection and image restoration.Secondly, to deal with the deficiency of switching filter, this paper presents a cloud model filter which is based on the adaptive MMEM filter. The cloud model filter combines a noise detector and a weighted fuzzy mean filter, which are both based on the cloud model. To eliminate the blindness in noise detection, the detector uses a cloud model to identify the noise from the uncorrupted pixels, according to their digital features. At the same time, it uses the uncertainties of the drops to make the membership degrees of the pixels "soft". Then, the filter describes the contributions of the pixels in image restoration by using the uncertainties of the drops. Based on this, it shows the pixels with different gray values will have different contributions in image restoration. The experimental results showed that, the cloud model filter was more effectiveness and universality than the adaptive MMEM filter. Even at a noise level as high as95%, the cloud model filter still can restore the image with good detail-preservation.Finally, paper presents a two dimensional cloud model filter which is based on the cloud model filter. The proposed filter restores the noise image by using a two dimensional cloud model filter with combining the gray values and the location information of the pixels, and shows image anisotropy in the processing. The experimental results showed that, compared with the cloud model filter, although the proposed filter was slightly lower in computational efficiency, however, preserved more image details in high noise levels, especially in the activity region of the image.