基于鲁棒Otsu的红外无损检测缺陷分割算法

Algorithm for defect segmentation in infrared nondestructive testing based on robust Otsu

在红外无损检测获取的图像中,缺陷区域与非缺陷区域所占面积比例悬殊,且图像经过序列增强处理之后仍然存在阴暗区域,导致缺陷分割准确性受损。为此,结合局部阈值分割法的相对阈值思想,提出一种基于鲁棒Otsu的缺陷分割算法。首先,引入邻域均值与邻域总梯度作为表征像素点的所属类别与空间状态的重要参数。然后,采用基于像素点-块区的统计调整模型对红外图像缺陷区和非缺陷区的灰度值进行动态调整。最后,采用基于灰度-邻域偏差的改进二维直方图及其区域划分方法,通过自动选取邻域边长的遗传算法搜索最佳阈值,实现红外图像的缺陷分割。结果表明:该算法不仅改善了Otsu算法的鲁棒性,且能够提高红外无损检测缺陷分割的准确性。

In infrared nondestructive testing, the proportion of defects is very different from that of background, and the low contrast region of infrared image has not been completely eliminated after image sequence enhancement, resulting in impaired accuracy of defect segmentation. In order to solve this problem, a defect segmentation method based on robust Otsu algorithm was proposed, which combined the relative threshold idea of local threshold segmentation method. Firstly, the mean value and the total gradient of the neighborhood were used to represent the category and spatial state of the pixels. Secondly, a point-block fusion statistical adjusted model on this basis was established for dynamically adjusting the gray scale values of the infrared image defects and non-defect regions. Finally, the improved two-dimensional histogram and its region division method based on gray value and neighborhood gray deviation was set for calculation of fitness function in genetic algorithm through which the optimal threshold could be determined from the mutative neighborhood size, then segmentation of defects could be achieved. The results show that this method improves the robustness of Otsu and the accuracy of defect segmentation.