Abstract

The article deals with the detection of document machine-readable zones (MRZ) in images obtained with the aid of small-size digital cameras. The herein proposed method is based on the mutual arrangement of binarized image connected components. A graph is plotted the nodes of which are the center of the black connected components. Distribution of the graph edges provides information on the document orientation whereby the algorithm is made rotation-invariant. Paths which satisfy special requirements and highly likely correspond to the MRZ lines are searched for in the graph. Such paths are clustered and then the most consistent cluster is selected with due regard for knowledge on possible MRZ geometrical characteristics. The square enclosing this cluster is the answer to the algorithm. Tests performed on open sets of data showed substantial improvement in detection quality as compared with the state-of-the-art methods. The computational complexity of the algorithm allows its real-time use in mobile devices.

中文翻译：

---

移动设备的摄像头捕获的图像中的机器可读区域检测

摘要

本文介绍了在借助小型数码相机获得的图像中对文档机器可读区域（MRZ）的检测。本文提出的方法基于二值化图像连接的组件的相互布置。绘制了一个图形，其节点是黑色连接的组件的中心。图形边缘的分布提供了有关文档方向的信息，从而使算法旋转不变。在图中搜索满足特殊要求并且极有可能与MRZ线相对应的路径。对这些路径进行聚类，然后在适当考虑可能的MRZ几何特征知识的情况下选择最一致的聚类。包围该簇的正方形是该算法的答案。与最新方法相比，对开放数据集进行的测试显示出检测质量的显着提高。该算法的计算复杂度允许其在移动设备中的实时使用。