The paper describes an algorithm developed to enhance the analysis (manual or automatic) of radiographic images acquired for samples of complex geometry (such as those enabled by additive manufacturing techniques), where the imprint of the sample’s geometric complexity in the radiograph is likely to undermine the ability to identify defect indications. The underlying premise is that, assuming the sample geometry is known (at least the CAD specification), a simulation of the experimental radiograph can be used to essentially subtract out the geometric complexity from an experimental image, revealing the deviations from the expected inspection output. The approach is especially relevant when the uniqueness of the sample (for example due to personal customisation) limits the availability of comparable experimental data. However, in practice, this technique requires the simulation to be accurately calibrated to the experimental configuration, necessitating the use of a numerical optimisation to fit the simulation parameters. As a by-product, the parameters of an imperfectly specified experimental set-up are recovered. The algorithm architecture described can operate on multiple input radiographs simultaneously, and is readily adaptable to other image-based inspection modalities. Results for several test inputs are presented, starting with synthetic test cases and ending with a set of three experimental radiographs. The results are convincing, as a difference image enables a substantial reduction in image bit-depth, making deviations of interest more apparent and demonstrating the value of the approach.

中文翻译：

---

基于仿真的复杂放射影像分析

该论文描述了一种算法，该算法旨在增强对复杂几何形状样本（例如通过增材制造技术实现的样本）采集的射线照相图像的分析（手动或自动），其中射线照相中样品几何复杂性的印记可能会破坏识别缺陷迹象的能力。基本前提是，假设样品几何形状是已知的（至少是 CAD 规范），可以使用实验射线照片的模拟来基本上从实验图像中减去几何复杂性，从而揭示与预期检查输出的偏差。当样本的独特性（例如由于个人定制）限制了可比实验数据的可用性时，该方法尤其重要。然而，在实践中，这种技术需要根据实验配置准确校准模拟，因此需要使用数值优化来拟合模拟参数。作为副产品，不完全指定的实验装置的参数被恢复。所描述的算法架构可以同时在多个输入射线照片上运行，并且很容易适应其他基于图像的检查方式。给出了几个测试输入的结果，从合成测试案例开始，以一组三张实验射线照片结束。结果令人信服，因为差异图像可以显着减少图像位深度，使感兴趣的偏差更加明显并证明该方法的价值。