基于多尺度降采样规范化割的图像裂纹检测  

Image crack detection with multi-scale down-sampled normalized cut

作　　者：王森[1] 伍星[1] 张印辉[1] 陈庆[1] 

机构地区：[1]昆明理工大学机电工程学院,昆明650500

出　　处：《仪器仪表学报》2017年第11期2788-2796,共9页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61461022;61761024)项目资助

摘　　要：针对多尺度规范化割在边缘检测时精度低以及求解特征向量耗时长等缺陷,提出一种基于多尺度降采样规范化割的图像裂纹检测方法。该方法首先利用反对称双正交小波变换的半重构特性对待测图像的多个尺度进行边缘特征提取;其次结合各尺度的强度和位置特征构建多尺度相似矩阵和多尺度规范化相似矩阵;然后对多尺度相似矩阵进行降采样并利用谱分割方法实现降采样特征向量求解;最后利用多尺度规范化相似矩阵对降采样特征向量进行上采样的乘法运算并离散化后得到最终结果。在3个数据集的单一目标图像上进行文中方法与多尺度规范化割等方法的实验结果表明,不仅提高检测精度,而且减少运算时间。An image crack detection method with multi-scale down-sampled normalized cut is proposed to address the problems of low precision for edge detection and time-consuming for feature vector solution with the muki-scale normalized cut. Firstly, the feature of half-reconstruction of the anti-symmetrical bi-orthogonal wavelet is used to extract the multi-scale edge features of the test image Then, combining the strength feature and location feature of each scale, the multi-scale similarity matrix and multi-scale normalized similarity matrix are obtained. Spectral segmentation method is utilized to calculate the down-sampled feature vector of the multi-scale similarity matrix after down-sampling. Finally, by multiplying the multi-scale normalized similarity matrix with the down-sampled feature vectors, the segmentation result is obtained after discretization. Experimental results indicate that the proposed method improves the accuracy of detection and reduces the computational time on three image datasets, compared with other methods.

关 键 词：反对称双正交小波变换 边缘检测 多尺度降采样规范化割 多尺度规范化相似矩阵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TH14[自动化与计算机技术—计算机科学与技术]