基于K-SVD算法的带钢表面缺陷图像去噪

发布时间：2018-02-16 21:15

本文关键词： K-SVD算法正交匹配追踪 DCT字典高斯噪声滤波带钢缺陷　出处：《表面技术》2017年05期 　论文类型：期刊论文

【摘要】：目的有效滤除带钢表面缺陷图像高斯噪声。方法高斯噪声是影响带钢图像质量的主要噪声类型之一,针对带钢表面缺陷图像高斯噪声去噪,首先对传统K-SVD(K-means and singular value decomposition)算法中的字典进行升级改造,然后采用正交匹配追踪(OMP,Orthogonal Matching Pursuit)算法对图像进行重构,滤除噪声,最后运用此算法对缺陷图像进行高斯滤波处理。为验证该算法去噪效果,选取几种常见的典型缺陷图像(划伤、气泡、氧化色、粘结纹)进行测试仿真,并选用中值滤波、均值滤波、小波变换、维纳滤波、3维块匹配(BM3D)等多种传统滤波方法进行比较。结果该算法对四种典型缺陷去噪的PSNR(Peak Signal to Noise Ratio)值平均可达33.976 d B,MSE(Mean Square Error)平均值为27.607,SSIM(Structural Similarity)平均值为0.912。结论该算法对带钢表面缺陷重构图像的边缘细节清晰,PSNR、MSE、SSIM三个性能指标明显优于其他传统滤波算法,去噪效果良好。
[Abstract]:Objective to effectively filter Gao Si noise from strip surface defect image. Methods Gao Si noise is one of the main noise types affecting the quality of strip surface defect image. Firstly, the dictionary in the traditional K-SVDmeans K-means and singular value decompositionalgorithm is upgraded, and then the orthogonal Matching pursuit algorithm is used to reconstruct the image and filter the noise. Finally, the algorithm is used to process the defect image by Gao Si filter. In order to verify the denoising effect of the algorithm, several common typical defect images (scratching, bubble, oxidizing color, bond pattern) are selected for test and simulation, and median filter is selected. Mean filter, wavelet transform, Several traditional filtering methods, such as Wiener filter and 3D block matched BM3D, are compared. Results the average value of PSNR(Peak Signal to Noise error of four typical defects can reach 33.976 d BMSE mean Square error (mean value is 27.607). Conclusion this algorithm has a structural Similarity0.912. The edge details of the reconstructed image of strip surface defects are clear and the three performance indexes of PSNRX MSESSIM are obviously superior to those of other traditional filtering algorithms. The denoising effect is good.
【作者单位】：河北工业大学电子信息工程学院;华北理工大学信息工程学院;哈尔滨工业大学电子与信息工程学院;
【基金】：河北省自然科学基金(E2016202341) 河北省引进留学人员基金(C2012003038)~~
【分类号】：TG142.1+1;;TP391.41

【参考文献】