论文部分内容阅读
在现有的X射线数字图像自动识别方法中,多采用对单幅数字图像进行孤立评判的方法。由于此类方法中阈值选取难以最优化,因而存在一定的误判率。为了解决这一问题,提出了一种X射线数字图像自动识别新方法。该方法将识别过程分为两步:缺陷提取和缺陷跟踪。第一步利用传统方法在每幅图像中分离出潜在缺陷。这一步保证真缺陷能全部提取出来,而不考虑伪缺陷的数量。第二步力图找出同一试件不同图像中分离出的缺陷之间的相互关系。如果第一步某一图像中分离出的某一缺陷在其他图像中都找不到相对应的缺陷区域,就定义该缺陷为伪缺陷,也就是说,真缺陷在不同图像中必须满足一定的几何关系。多幅图像中的缺陷跟踪综合利用了极线约束、三维重建和三线性约束等立体视觉算法。该方法的检测效果已经利用航空发动机叶片X射线数字图像得到验证。试验结果表明:利用该方法可以提高真缺陷的识别率,降低误判率。
In the existing X-ray digital image automatic recognition method, the method of isolating a single digital image is mostly used. Due to the difficulty of optimizing the threshold selection in such methods, there is a certain misjudgment rate. In order to solve this problem, a new method of X-ray digital image automatic recognition is proposed. This method divides the recognition process into two steps: defect extraction and defect tracking. The first step uses traditional methods to isolate potential defects in each image. This step to ensure that all real defects can be extracted, without regard to the number of false defects. The second step seeks to find the same specimen in different images of the separation between the defects of the interrelationship. If one of the defects separated in a certain image of the first step can not find the corresponding defect area in other images, the defect is defined as a false defect, that is to say, the true defect must meet certain values in different images Geometric relationship. Defect tracking in multiple images integrated the stereo vision algorithms such as polar constraint, 3D reconstruction and trilinear constraint. The test results of this method have been validated using X-ray digital images of aeroengine blades. The experimental results show that this method can improve the recognition rate of true defects and reduce the false positive rate.