Research On Image Denoising Algorithm Based On Local Hu Moment

Image denoising is an important preprocessing step in the field of digital image processing.The effect of its treatment will directly affect the follow-up work.Noise is ubiquitous,it may not only affect the image of their own quality,but also give us the wrong message,so the image denoising is quite important.The traditional image denoising is based on local,although achieved some effect,but will make the edge blurred.And the local mean denoising breaks through the local shackles.By using the self-similarity of the image,the comparison of the pixels becomes the comparison between the neighborhood blocks in the image.The comparison of neighborhood blocks is more stable than the comparison between points,so the similarity between the pixel and the pixel is estimated by measuring the similarity between the neighborhood blocks,and then the result of denoising is obtained by averaging these points to obtain a better denoising effect and to maintain the edge of the structural information.However,the classical nonlocal mean denoising method also has some drawbacks because it uses a weighted Euclidean distance to measure the similarity between neighborhood blocks,while the Euclidean distance is not strong and is susceptible to noise,so the Euclidean distance can not accurately measure the similarity between neighborhood blocks,thus affecting the effect of-denoising.In this paper,nonlocal mean denoising algorithm based on local Hu moments is proposed to solve the problem of block similarity measure by using Euclidean distance in classical nonresidential denoising algorithm.From the aspects of anti-noise,structural characteristic matching and energy independence,improve the performance of the block similarity measure,so as to improve the algorithm's denoising effect as a whole.The local Hu moments are constructed by combining the weight function of the krawtchouk polynomial with the image function to construct a new weight function of the geometric moment to obtain a new geometric moment.We use the new geometric moment of the structure to obtain a new central moment.The second and third order moments form seven invariant moments to form a set of eigenvectors,this set of feature vectors has not noly rotation,scale,translation invariance,but also strong anti-noise ability.The similarity of neighborhood feature vectors is measured from higher dimensions,and then it is used as part of the weights to measure neighborhood similarity.By comparing and analyzing the Gaussian white noise images and salt and pepper noise images with different intensities,we can see that the block similarity based on local Hu moments is superior to the traditional Euclidean distance comparison.In this paper,the non-local mean denoising algorithm based on local Hu moment is improved obviously in the peak signal-to-noise ratio and structure similarity to the classical algorithm,and the structure information of the image can be better preserved.