全文下载：
202003013.pdf
文章编号： 1672-6987（2020）03-0092-12； DOI： 10.16351/j.1672-6987.2020.03.013
王晨， 贾晓芬*
（安徽理工大学 电气与信息工程学院，安徽 淮南 232001）
摘要： 针对传统图像去噪方法出现纹理细节模糊的问题，提出一种利用误差多样性集成子SVM的图像椒盐噪声去除方法SVM-WCEC。首先，在训练图像上移动3×3窗口，提取每个窗口中心像素点的局部二值和加权差分特征作为输入训练子SVM，利用加权计数值对各子SVM投票，选择得票数最多的子SVM组合作为分类器模型；然后，用相同特征提取方法遍历含噪图像提取特征输入分类器模型，将像素点分为噪点和信号点；最后，在3×3滤波窗口内，用非线性的权重均值滤波估计噪点灰度值，和直接输出的信号点灰度值重构得到去噪后图像。在图像集BSD68上实验结果表明：与现有先进方法DAMF相比，SVM-WCEC的平均PSNR/SSIM值提高了1808 0dB/0150 4。实验数据充分说明：SVM-WCEC在去噪同时能很好地保留图像的纹理信息，获得较高的PSNR、SSIM和更好的视觉效果。
关键词： 图像去噪； 椒盐噪声； 集成学习； 支持向量机； 特征提取
中图分类号： TP 391文献标志码： A
引用格式： 王晨， 贾晓芬. 集成SVM的图像椒盐噪声去除方法\[J\]. 青岛科技大学学报（自然科学版）， 2020， 41（3）： 92-103.
WANG Chen, JIA Xiaofen. Image salt and pepper noise removal method based on SVM ensemble\[J\]. Journal of Qingdao University of Science and Technology（Natural Science Edition), 2020， 41（3）： 92-103.
Image Salt and Pepper Noise Removal Method Based on SVM Ensemble
WANG Chen, JIA Xiaofen
(School of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan 232001, China)
Abstract: Aiming at the problem that texture details are blurred in traditional image denoising methods, an image salt and pepper noise removal method based on error diversity ensemble sub-SVM (support vector machine) is proposed. Firstly, a 3×3 window is moved on the training image to extract the local binary and weighted difference features of the central pixels of each window as the input training sub-SVM, and the weighted counting values are used to vote on each sub-SVM, and the sub-SVM combination with the largest number of votes is selected as the classifier model. Then, the noisy image is traversed to extract the feature of the central pixel by the same feature extraction method and then input into the classifier model, and the pixels are divided into noise points and signal points. Finally, in the 3×3 filtering window, the gray value of the noise points is estimated by the non-linear weighted mean filter, the denoised image is reconstructed by the estimated the gray value of noise pixels and the gray value of signal pixels which are directly outputted. The experimental results on image set BSD68 show that, compared with the existing advanced method DAMF, the average PSNR/SSIM values of SVM-WCEC is increased by 1808 0 dB/0150 4. The experimental data fully shows that SVM-WCEC can preserve the texture information of the image while denoising, and obtain higher PSNR, SSIM and better visual effect.
Key words: image denoising; salt and pepper noise; ensemble learning; support vector machine; feature extraction
收稿日期： 2019-08-18
基金项目： 国家自然科学基金项目(61501006)；安徽高校自然科学研究重大项目(KJ2018ZD008)；安徽高等学校自然科学研究重点项目(KJ2017A076).
作者简介： 王晨（1995—），女，硕士研究生.*通信联系人.
