Typical crack detection processes in digital images produce a binary-segmented image that constitutes the basis for all of the following analyses. Binary images are, however, an unsatisfactory data format for advanced crack analysis algorithms due to their sparse nature and lack of significant data structuring. Therefore, this work instead proposes a new approach based on Euclidean graphs as functional crack pattern descriptors for all post-detection analyses. Conveying both geometrical and topological information in an integrated representation, Euclidean graphs are an ideal structure for efficient crack path description, as they precisely locate the cracks on the original image and capture salient crack skeleton features. Several Euclidean graph-based algorithms for autonomous crack refining, correlation and analysis are described, with significant advantages in both their capabilities and implementation convenience over the traditional, binary image-based approach. Moreover, Euclidean graphs allow the autonomous selection of specific cracks or crack parts based on objective criteria. Well-known performance metrics, namely precision, recall, intersection over union and F1-score, have been adapted for use with Euclidean graphs. The automated generation of Euclidean graphs from binary-segmented images is also reported, enabling the application of this technique to most existing detection methods (e.g., threshold-based or neural network-based) for cracks and other curvilinear features in digital images.

中文翻译：

---

欧几里得图作为裂纹模式描述符用于数字图像中的自动裂纹分析

数字图像中的典型裂纹检测过程会生成二进制分割图像，该图像构成以下所有分析的基础。然而，二进制图像由于其稀疏性和缺乏重要的数据结构，对于高级裂纹分析算法来说是一种不能令人满意的数据格式。因此，这项工作提出了一种基于欧几里得图的新方法，作为所有检测后分析的功能裂纹模式描述符。欧几里得图以集成表示形式传达几何和拓扑信息，是有效裂纹路径描述的理想结构，因为它们可以精确定位原始图像上的裂纹并捕获显着的裂纹骨架特征。描述了几种基于欧几里得图的自主裂纹细化、相关和分析算法，与传统的基于二进制图像的方法相比，它们在功能和实施便利性方面具有显着优势。此外，欧几里得图允许基于客观标准自主选择特定裂缝或裂缝部分。众所周知的性能指标，即精度、召回、联合交集和 F1 分数，已适用于欧几里得图。还报告了从二进制分割图像自动生成欧几里得图，使该技术能够应用于大多数现有的检测方法（例如，基于阈值或基于神经网络），以检测数字图像中的裂缝和其他曲线特征。欧几里得图允许基于客观标准自主选择特定裂缝或裂缝部分。众所周知的性能指标，即精度、召回、联合交集和 F1 分数，已适用于欧几里得图。还报告了从二进制分割图像自动生成欧几里得图，使该技术能够应用于大多数现有的检测方法（例如，基于阈值或基于神经网络），以检测数字图像中的裂缝和其他曲线特征。欧几里得图允许基于客观标准自主选择特定裂缝或裂缝部分。众所周知的性能指标，即精度、召回、联合交集和 F1 分数，已适用于欧几里得图。还报告了从二进制分割图像自动生成欧几里得图，使该技术能够应用于大多数现有的检测方法（例如，基于阈值或基于神经网络），以检测数字图像中的裂缝和其他曲线特征。