基于显著性判别的缺陷检测方法

X射线成像存在噪声大、半影现象及散射等问题，使所得缺陷图像边缘模糊，背景灰度变化不均匀，严重影响了缺陷的识别准确率。提出了一种由LoG边缘检测和局部对比度筛选进行显著性判别的缺陷检测方法，在基于LoG边缘检测的双阈值进行显著性边缘检测的基础上，通过各向同性扩散方法求取待定缺陷的局部背景，利用待定缺陷和局部背景之间对比度的显著性设置第三个阈值进行进一步判断，从而去除假缺陷，使缺陷能够被准确地提取出来，并可以同时确定缺陷的轮廓和面积。实验结果表明，该方法对缺陷识别的准确度较高，并可以用于在线的实时检测系统中。

A defect detection method based on saliency discrimination

X ray imaging has many problems such as noise, penumbra and scattering, which results in blurred edge of the image and the uneven background gray level and seriously affects the accuracy of defect identification. We propose a new defect detection method for X ray images based on LoG edge detection and local contrast to carry out saliency discrimination. On the basis of the salient edge detection based on the double thresholds of LoG edge detection, the local background of undetermined defects is obtained by the isotropic diffusion method. We also set a third threshold by utilizing the contrast saliency between undetermined defects and the local background to remove the false defects, so that the defects can be accurately extracted and the contour and area of defects can be determined at the same time. Experimental results show that the algorithm has high accuracy for defect identification and can be used for online real-time detection systems.