The Research Of Non-local Means Filtering To Remove The Image Gaussian Noise

The image obtained from the real world is always mixed with the noise, which comes mainly from the process of capturing images and transferring images. Since the noise greatly reduced image detail and makes images difficult to read, finding an appropriate method to remove or reduce noise is very necessary. Non-local means algorithm provides a powerful framework for image denoising.It takes advantage of the similarity feature in the entire image which means some details of the partial image will appear in many times. Using non-local self-similarity to suppress image noise, this denoising algorithm is a very famous.Although it can achieve a good denoising effect, there are some drawbacks depressed the denoising results. It takes a lot of time to compute the similarity weights and is difficult to find the candidate pixel from different patches. Calculating the similarity between the image patches is the key technology in NLM algorithm.When the traditional block-matching algorithm calculated the similarity weight between the two image patches, it didn't take into account the angle of rotation between the patches. This paper calculated the rotation angle of the patch on the base of the main direction of the corner and combined with the traditional block matching to calculate the similarity. This method avoided the similarity weight error caused by the rotation and the similarity weight was more accurate than before. Experiments showed that this method retained more image details than the original model and get a better noise cancellation.Curvelet transform was a multi-resolution,localized, multi-directional image representation. The Nonlocal-means filter based on fast curvelet transform defined the reconstructed image at different levels, and determined the similarity weights between noisy image patches. The reconstructed image contained the entire image feature, which can be estimated by the standard deviation of noise noisy image and obtained from different levels noisy images converted by curvelet transform.Curvelet widely applied to image reconstruction because curvelet reconstruction demonstrated a higher perceived quality than wavelet reconstruction. The denoised image had a good visual effect, high quality edge information restoration, and a slight linear and curvilinear feature.