基于轮廓波包变换的磁瓦表面缺陷提取

Defect Extraction on Magnetic Tile Surface Based on Contourlet Packet Transform

针对磁瓦图像对比度低、背景复杂及亮度不均匀等特点,结合磁瓦表面缺陷采用机器视觉自动检测的需求,提出一种基于平稳小波包和非下采样方向滤波器组构造的轮廓波包变换的缺陷提取方法.首先详细论述了轮廓波包变换构造原理,利用平稳小波包变换时不变特性,建立一种新的轮廓波包变换来保证信号的多尺度方向选择性.其次,利用轮廓波包系数的相关特性,采用相关去噪算法消除磁瓦缺陷图像噪声;根据轮廓波包子带系数在不同区域的三维特性,使用自适应阈值修正子带系数,保留缺陷子带系数.最后,采用轮廓波包逆变换重构轮廓波包子带系数,获得缺陷图像.实验结果表明,相比于Sobel和小波包的边缘检测方法,该方法能有效消除磁瓦图像噪声,提取缺陷图像准确率可达95%.

To meet the needs of the machine vision automatic detection system about the magnetic tile surface defects,a defect extraction method is proposed in this paper based on the Contourlet packet transform( CPT) formed by the stationary wavelet packet( SWP) and nonsubsampled directional filter bank( NSDFB). This method fully considers the low contrast,complex background and uneven brightness of the magnetic tile surface defects. Through analyzing the CPT theory and the SWP time-invariant feature,a new Contourlet packet transform is created to ensure the signal multi-scale direction selectivity. According to the signal CPT coefficients with high inter-scale correlations,the image denoising algorithm based on CPT is proposed. As the CPT sub-band coefficients have different 3D features,the adaptive threshold can fit these coefficients and make the defect coefficients retained. Then the new CPT coefficients are reconstructed via the inverse Contourlet packet transform,and the magnetic tile surface defects are got. The experimental results indicate that the proposed method can eliminate the magnetic tile image noise effectively and improve the accuracy rate of extracting the defects to 95% when it is compared with the Sobel and wavelet packet.