The Research On Image Denoising Method Based On Image Singularity Of Multi-scale Transform

With the increasing popularity of image information, the images have a very important role in many areas. But the images often subject to different noise and different degrees of pollution in the collection, access and transmission process. In order to follow up to a higher level of treatment, it is essential to use denoising pre-processing to the image.With the rapid development of wavelet theory and its good time-frequency localization properties, the wavelet theory provides a good tool in the process of reducing noise while protecting image details. The wavelet transform can describe the signal of non-stationary features primly, such as: edges, peak, breakpoints and so on, which facilitate the extraction of image features. In addition, wavelet transform also has a low entropy nature, relevant nature, multi-resolution, flexible base and so on. With using these features adequately, we can distinguish between noise and signal well in the wavelet transform domain.The curvelet transform not only has the multi-scale features, but also has the anisotropic characteristics because of the introduction of an orientation parameter. And its anisotropic characteristics make it have a good characterization to the line features. For the two-dimensional image, the line feature contains edges, contour lines and so on just contains the most important image information, which makes curvelet transform have good performance in the field of digital image processing.Image denoising is the basic problems in the image processing and analysis, how to maintain more image information while removing more noise has become important problem to researchers. At present, wavelet transform and curvelet transform theory have been noticed widely because they has unique advantages in analyzing the image singularity. This paper's main-work is as follows:(1) Based on the wavelet threshold denoising method, we adjusted the wavelet coefficients. In this proposed approach, we adjusted the high-frequency coefficients whose absolute value is smaller than the threshold and biggest in the high-frequency coefficients and maintain the high-frequency coefficients whose absolute value is bigger than the threshold in the first step, then using wavelet image denoising method to the adjusted high-frequency coefficients. In this proposed method the adjusted percent correlates to the high-frequency coefficients and the threshold.(2) In the traditional image denoising method, the image edge information is not taken into account primly while removing the image noise, so the denosing image has a certain degree of edge blur. In this paper, we propose an image denoising method based on image edge protection. This method extracts image edge information in the first step, and then protects the information. The last step is using the traditional image denoising method to the disposed image.(3) Based on the wavelet and curvelet transform have unique advantages in disposing point singularity and curve singularity respectively, we proposed a new image denoising method based wavelet and curvelet transform. In this method, we use wavelet transform and curvelet transform to the noise image partly, and then reconstruct the image based on the weight coefficients of wavelet transform and curvelet transform. In this proposed approach, the weight coefficients is related to the image singularity.