文献类型：期刊文献

英文题名：A noise level estimation method of impulse noise image based on local similarity

作者：Lin, Cong[1,2];Ye, Youqiang[2];Feng, Siling[1];Huang, Mengxing[1,3]

机构：[1]Hainan Univ, Coll Informat & Commun Engn, Haikou 570228, Hainan, Peoples R China;[2]Guangdong Ocean Univ, Coll Elect & Informat Engn, Zhanjiang 524000, Peoples R China;[3]State Key Lab Marine Resource Utilizat South Chin, Haikou 570228, Hainan, Peoples R China

年份：2022

卷号：81

期号：11

起止页码：15947

外文期刊名：MULTIMEDIA TOOLS AND APPLICATIONS

收录：SCI-EXPANDED(收录号：WOS:000763256600023)、、EI(收录号：20221011741390)、Scopus(收录号：2-s2.0-85125538884)、WOS

基金：This work was supported by National Key Research and Development Program of China (Grant #: 2018YFB1404400), Hainan Provincial Natural Science Foundation of China (Grant #: 2019CXTD400).

语种：英文

外文关键词：Noise level estimation; Local similarity; Random-value impulse noise

外文摘要：The detection and removal methods of impulse noise often need to estimate the noise level of the damaged image in advance to obtain a better detection rate. An effective method of random-value impulse noise level estimation based on local similarity is proposed in this paper. Firstly, quantify the similarity between the center pixel x and any pixel y in the neighborhood of x based on their Euclidean distance and gray difference, then the similarity of pixel x and each pixel in its neighborhood is accumulated and summed to obtain the local similarity (LS) of pixel x. The value of LS represents the local consistency of a given pixel with respect to its neighboring pixels in which can also determine if a pixel is an impulse noise or a clean pixel. Hence, the LS value of a pixel could be regarded as an effective index to measure whether it is a clean pixel. Then the noise pixels in multiple flat regions of the noise image are detected to obtain the noise level of each region, and the noise level of these flat regions are processed with average operation to estimate the impulse noise level of the entire image finally. Extensive experiments were conducted to verify the effectiveness of the method and the experimental results show that the method is effective in scenarios with various noise levels, and the estimation error of the noise level of most images is within 1%. By comparing the RMSE and Std of different noise level estimation algorithms, it can be found that the algorithm proposed in this paper has higher robustness and accuracy, which can be well applied to practical applications with impulse noise level as the key parameter.