基于多尺度降采样规范化割的图像裂纹检测

发布时间：2018-04-02 16:10

本文选题：反对称双正交小波变换　切入点：边缘检测　出处：《仪器仪表学报》2017年11期

【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。
[Abstract]:According to the multi-scale normalized cut edge detection accuracy is low, eigenvector time-consuming defect is proposed based on a multi-scale reduced image crack detection method standardized cut sampling. Using the method of multiple scales antisymmetric to the image reconstruction characteristics of semi biorthogonal wavelet transform for edge feature extraction; secondly, constructing multiscale similarity matrix and multi-scale normalized similarity matrix combined with the intensity and position characteristics of different scale; and then reduces the sampling and spectral segmentation methods down sampling feature vector to solve multiscale similarity matrix; finally, using multi-scale normalized similarity matrix multiplication of down sampling feature vectors for sampling and after discretization of the single target image to get the final result. In the 3 sets of data on the method and multiscale normalized cut method and the experimental results show that, It not only improves the detection precision, but also reduces the operation time.

【作者单位】：昆明理工大学机电工程学院;
【基金】：国家自然科学基金(61461022,61761024)项目资助
【分类号】：TP391.41

【相似文献】