Noise Level Estimation Algorithms Based On Fast Local Means And Its Application

With the development of modern digital information technology,image has been an indispensable way to get information for people in recent years.Since the equipments used to handle images are not perfect,images are inevitably affected by noise during the process of acquisition,transmission and storage.If the image contains noise,it is difficult for people to receive image information and the subsequent processing of the image will also be affected.Therefore,image denoising is still challenge in image research area.To date,most of the existing image denoising algorithms need to provide the image noise level parameter before dealing with the noise of image.Hence,it needs to use an efficient and accurate noise estimation algorithm to estimate the noise level of image.Nevertheless,the traditional noise level estimation algorithm still has many drawbacks:(1)the estimation effect of algorithm is unsatisfactory when the noise ratio of image is higher.(2)The ability of anti-interference for the algorithm is not good.Furthermore,the estimation effect of algorithm will be severely affected if the image contains other types of noise.(3)Most of estimation algorithms can only to estimate some special noise.In order to avoid the shortcomings of the traditional noise estimation algorithms,we proposed a novel noise level estimation algorithm based on fast local means called FLM-NLE(Noise Level Estimation Based on Fast Local Means).FLM-NLE utilizes the principle that similar noise image usually has similar statistical property to estimate the noise level of image by weighted scheme.Specifically,some widely representative and clean images were selected and corrupted by different types of noises with different variances for constituting a set of distorted images.And image features was extracted by BRISQUE method(the features extracted by BRISQUE are natural scene statistical characteristics and are sensitive to image distortion.Therefore,the features are not only a very good description of noise,but also the speed of extracting feature of image is fast).Then,the distortion feature vector library was established by feature vector of images and the corresponding noise variance values.Finally,the distortion feature vector library was divided into several clusters feature vector library by clustering to speed up the subsequent noise estimation.During the noise level estimation,Firstly,the feature vector of an image to be estimated was extracted by the same feature extraction approach.Secondly,the feature vector of the image to be estimated is assigned to a cluster using smallest distance criteria.In the end,we use the local means algorithm to estimate the noise level parameter of image based on feature vectors belong to the cluster.To further verify the validity of the proposed FLM-NLE algorithm,this paper carried out many experiments with different standard images.And the experimental results show that the FLM-NLE algorithm can accurately and quickly estimate the noise level for any given noise image.In addition,FLM-NLE algorithm behave robustly and the algorithm can be applied to estimate many types of noise.In order to solve the problem that the noise level parameter of image denoising algorithm needs to be manually set in actual use,the FLM-NLE algorithm is introduced to the BM3 D image denoising algorithm(Block-matching and 3D Filtering),that is an improved BM3 D filter—IBM3D.And the experimental results show that the noise denoising effects of IBM3 D coincide with the BM3 D when the noise level parameter is set the growed-truth.