In recent years, digital copies of paper documents are used widely with the prevalence of various online services. As a result, it is critical to validate the authenticity of the uploaded document images to protect against attacks from malicious users. Out of various types of attacks, the recapture attack (by reprinting and recapturing) is effective in concealing the trace of document forgeries. However, detecting the recaptured document images is challenging. To address this problem, we first study the halftone cell distortion introduced in both the genuine and recaptured document images. Based on our study, a unified model that characterizes the halftone cell distortion (e.g., errors in size and displacement) is then proposed for accurate estimation of the distortion parameters. The statistics of the estimated parameters are then exploited in a hypothesis testing framework to detect the recaptured document images. The questioned document image can be authenticated by testing against the null hypothesis, i.e., the image is a genuine sample. To evaluate the performance of the proposed approach under different application scenarios, extensive experiments are conducted with different prior knowledge of printers (known printer model, known printing technique, and In-The-Wild (unknown printing device and document contents)). The experiment results show that the proposed approach outperforms the data-driven benchmark approaches by a significant margin. Specifically, under the In-The-Wild experiment protocol, the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) of the proposed approach is above 0.87 while the AUC of the benchmark approaches (even some utilize both genuine and recaptured samples) degrades to less than 0.77.

中文翻译：

---

基于半色调单元统一失真模型的文档重抓检测

近年来，随着各种在线服务的普及，纸质文档的数字副本被广泛使用。因此，验证上传的文档图像的真实性以防止恶意用户的攻击至关重要。在各种类型的攻击中，重新捕获攻击（通过重新打印和重新捕获）可以有效地隐藏文件伪造的痕迹。然而，检测重新捕获的文档图像具有挑战性。为了解决这个问题，我们首先研究了在真实和重新捕获的文档图像中引入的半色调单元失真。基于我们的研究，然后提出了一个表征半色调单元失真（例如，尺寸和位移误差）的统一模型，用于准确估计失真参数。然后在假设检验框架中利用估计参数的统计数据来检测重新捕获的文档图像。可以通过针对零假设的测试来验证被质疑的文档图像，即图像是真实的样本。为了评估所提出的方法在不同应用场景下的性能，使用打印机的不同先验知识（已知的打印机型号、已知的打印技术和 In-The-Wild（未知的打印设备和文档内容））进行了广泛的实验。实验结果表明，所提出的方法明显优于数据驱动的基准方法。具体来说，在 In-The-Wild 实验协议下，所提出方法的接收器操作特性 (ROC) 曲线 (AUC) 下的面积大于 0。