基于引导滤波和NSST的工业CT图像边缘检测

针对传统边缘检测算法不能准确检测有噪工业CT图像边缘的问题，提出一种鲁棒性好、能有效保持细小边缘的边缘检测算法.用引导滤波取代高斯滤波作为边缘检测的预处理，避免Canny算法对边缘的损坏，得到初步检测结果.在此基础上采用非下采样Shearlet变换分解图像，提取包含图像边缘细节信息的各尺度不同方向的高频系数.对每个方向的系数进行模极大值检测，并结合不同分解程度下边缘像素处的系数关系进一步调整模极大值，低频置零并通过反变换得到高频边缘检测结果.将初步检测结果与高频检测结果进行融合，经数学形态学处理得到最终边缘检测图像.实验对比了Canny算子以及近年来提出的同类边缘检测算法的结果，所提算法表现出更好的边缘保持特性，检测的完整性和准确性更高，品质因数比实验中的其他算法平均高出12%，边缘检测效果优越，为工业CT无损检测系统提供了更好的边缘检测方案.

Edge Detection for Industrial CT Image Based on GuidedFiltering and Non-subsampled Shearlet Transform

As existing image edge detection algorithms cannot accurately detect edges from noisy industrial computed tomography (CT) images, a robust edge detection algorithm capable of preserving fine edges is proposed. Instead of Gaussian filtering, guided filtering is used in image pre-processing for edge detection to avoid edge destruction of the Canny algorithm. Having obtained the preliminary detection result, non-subsampled shearlet transform (NSST) is applied for image decomposition. High-frequency coefficients of various scales in different directions containing edges and details are extracted. Modulus maximum detection is performed on the coefficients in each direction, and the maximum modulus values are adjusted depending on the property of coefficients of the edge points under different decomposing conditions. By setting the low-frequency coefficients to zero, inverse NSST is performed to get the high-frequency edge detection result. Finally, the preliminary result and the high-frequency detection result are combined. The final edge map is obtained with mathematical morphology. Experiments are performed and detection results are compared with those of classical Canny algorithm and several recent and similar edge detection algorithms. The proposed algorithm shows better edge preserving property, higher edge integrity and accuracy. An average increase of 12% of the figure of merit (FOM) indicator is achieved. The proposed edge detection algorithm provides a better edge detection scheme for industrial CT nondestructive testing systems.