小波与中值滤波相结合的汽车管路连接件表面缺陷图像去噪

Despeckling for surface defect image of automotive pipe joints based on wavelet and median filter

由于汽车管路连接件表面缺陷图像中的高频噪声和椒盐斑状噪声直接影响缺陷特征提取的精度,提出一种小波与中值滤波相结合的去噪新方法。该方法首先对原始图像进行小波变换分解,分离出高频信号与高频噪声,并将噪声滤除,然后由灰度值变化曲线统计得到缺陷灰度分界阈值,结合中值滤波算法自适应地滤除椒盐斑状噪声以保护缺陷特征边缘,对去噪后的图像进行线性增强,使缺陷边缘轮廓更加清晰,最后采用Sobel边缘算子算法分别对中值滤波、小波滤波、高斯滤波及新方法去噪后的图像进行缺陷特征提取以对比分析去噪效果。实验结果表明,新方法的峰值信噪比(PSNR)分别比中值滤波、小波滤波及高斯滤波提高了10.70%、8.99%和8.87%;结构相似度(SSIM)分别提高了21.82%、23.34%和11.54%,说明新方法具有良好的去噪效果,并在一定程度上提高了缺陷细节形状特征提取的准确性,适用于汽车管路连接件表面缺陷的检测与分类识别。

High-frequency noise and salt-and-pepper spotty noise are widely present in the surface defect image of automotive pipe joints,these noises affect the accuracy of image defect’s feature extraction.Therefore,a new denoising method combining wavelet and median filtering was proposed.First,the original image was decomposed by wavelet transform to separate the useful high frequency signal and the high frequency noise,and the high frequency noise was filtered out. Then,according to the gray value change graph of the defect image,the defect boundary threshold could be obtained from the data statistics of the gray value of the defect position and the gray value of the non-defect position. Applying the threshold,the median filtering algorithm could adaptively filter out salt-and-pepper spotty noise while protecting the edge of the defect feature.The denoised image was linearly enhanced to make the defect edge clearer. Finally,the median filtering,wavelet filtering,Gaussian filtering and the new method proposed were used to denoise the image,the effects of the four denoising methods were compared and analyzed using the Sobel edge operator to extract the defect features of the four denoising images respectively.The following results can be gained.The Peak Signal to Noise Ratio(PSNR) of the new denoising method was 10.70 %,8.99 %,and 8.87 % higher than the median filtering,wavelet filtering and Gaussian filtering respectively.The Structural Similarity Index(SSIM) was increased by 21.82 %,23.34 %and 11.54 % respectively.These results indicate that the proposed method has good denoising effect and improves the accuracy of defect detail shape feature extraction in a way.It will be suitable for surface defect detection and classification identification of automotive pipe joints.