A novel local denoising scheme based on context

There are two types of traditional denoising methods: one is neighborhood method; the other is contextual method. Recently, some hybrids are proposed and reported good denoising results. However, the basic idea about these hybrids is the parameter of the image is estimated in a set of moving windows with the context, which leads to high complexity to the algorithms. Besides this, some moving windows cannot ensure the numbers of points that have the same context are large enough to obtain reliable estimated parameters. In this paper, we proposed a novel denoising scheme, which can adjust the sizes of local windows automatically according to the numbers of the contextual points. The division of the same contextual points is obtained by dividing the subband into four equal squares if the number of the points is in a suitable extension. Then the division can be done step by step until the number of the points is not in the extension. All divisions can be obtained according to these steps. The other assumption about our framework is the parameter in the same local window is same. Therefore, we can share statistical information among these pixels. Based on these assumptions, we propose a simple example to demonstrate the power of our new scheme. The experimental results show that the new framework improves the denoising results greatly even using the simplest model.