Data and Code Underlying the Bachelor Thesis: Binarization of Historical Watermarks

<strong>Data and Code Underlying the Publication: Binarization of Historical Watermarks</strong>These files provide the data and code used for the Bachelor's thesis, "Binarization of Historical Watermarks: A Review of Thresholding Techniques Applied to Historical Watermark Images." The objective of this thesis was to review the effectiveness of different thresholding techniques when applied to noisy historical watermark images. To this end, several different thresholding techniques were programmatically implemented. Data was collected for both a qualitative and quantitative evaluation, separately. For the qualitative evaluation, watermarks were randomly sampled from a private watermark dataset owned and provided by the German Museum of Books and Writing (https://www.dnb.de/EN/Ueber-uns/DBSM/dbsm_node.html##sprg315370), with their permission. These sampled watermarks were thresholded using several techniques. Consenting participants were then asked to fill out a survey regarding which technique they thought was most effective for each watermark image. For qualitative data, a dataset of human sketches [1] was randomly sampled, split into test and validation sets, and noised to appear like watermarks. The F1 Score, PSNR, NRM, and MPM metrics were calculated for each pair of clean and noised images.<br>Note that all participant data is anonymized, and no original watermark data is included due to copyright restrictions.<br><strong>Organization of the data</strong>- `code.zip`: This file contains all the code used during the research. This includes implementations of thresholding techniques, as well as code used for data processing. It should be noted that none of the watermark data, either real or synthetic, is included in this file. For this reason, some paths that lead to images in the code will not work.- `qualitative_data.zip`: This file contains the anonymized survey data filled in by participants, as well as the resulting files produced by processing the survey data. A copy of the original survey is not included, since permission has not been gained to redistribute the watermark images which are included in the survey. The code used to generate the processed results can be found in `code.zip`.- `quantitative_data.zip`: This file contains `.csv` files that detail the images sampled from the Human Drawings dataset [1], and the results after processing and evaluating these images. The code used to generate the processed results can be found in `code.zip`.<br><strong>References</strong>[1] M. Eitz, J. Hays, and M. Alexa, “How do humans sketch objects?” ACM Trans. Graph., vol. 31, no. 4, pp. 1–10, Aug. 2012, doi: 10.1145/2185520.2185540.

<strong>学术论文配套数据与代码：历史水印二值化</strong>本文件集提供了学士学位论文《历史水印二值化：应用于历史水印图像的阈值化技术（thresholding techniques）综述》所使用的数据与代码。本论文旨在评估各类阈值化技术针对带噪历史水印图像的应用效果。为此，研究团队通过编程实现了多种不同的阈值化技术，并分别采集了用于定性评估与定量评估的数据集。

对于定性评估，研究团队经德国书籍与写作博物馆（German Museum of Books and Writing，https://www.dnb.de/EN/Ueber-uns/DBSM/dbsm_node.html##sprg315370）授权，从其私有水印数据集内随机采样水印样本。使用多种阈值化技术对这些采样得到的水印图像进行二值化处理后，邀请经知情同意的参与者填写问卷，以评估他们认为针对每张水印图像效果最优的阈值化技术。此外，研究团队还从人类手绘草图数据集（Human Drawings dataset）[1]中随机采样样本，将其划分为测试集与验证集并添加噪声以模拟水印图像风格，用于定性评估的补充。针对每张干净图像与带噪图像对，研究团队计算了F1分数（F1 Score）、峰值信噪比（PSNR）、归一化均方误差（NRM）与平均像素匹配度（MPM）四项评价指标。

请注意，所有参与者数据均已匿名化处理；受版权限制，本数据集未包含原始水印数据。

<strong>数据组织方式</strong>
- `code.zip`：本文件包含研究过程中使用的全部代码，涵盖各类阈值化技术的实现代码与数据处理代码。需特别说明的是，本文件未包含任何真实或合成的水印数据，因此代码中部分指向图像文件的路径将无法正常工作。
- `qualitative_data.zip`：本文件包含参与者填写的匿名问卷数据，以及对问卷数据进行处理后得到的结果文件。本文件未包含原始问卷副本，原因是未获得问卷中所用水印图像的再分发授权。用于生成处理后结果的代码可在`code.zip`中找到。
- `quantitative_data.zip`：本文件包含若干`.csv`格式文件，详细记录了从人类手绘草图数据集（Human Drawings dataset）[1]中采样的图像，以及对这些图像进行处理与评估后得到的结果。用于生成处理后结果的代码可在`code.zip`中找到。

<strong>参考文献</strong>
[1] M. Eitz, J. Hays, and M. Alexa, "How do humans sketch objects?" ACM Trans. Graph., vol. 31, no. 4, pp. 1–10, Aug. 2012, doi: 10.1145/2185520.2185540.