AnonymousAuthors123/Halftonerevolution
收藏资源简介:
Halftone Revolution 是一个历史图像数据集,专为档案报纸图像的图像检索、相似性检测和视觉聚类研究而设计。数据集包含数字化的历史照片,主要来源于 Victor Forbin 收藏和 Rol 机构,这些作为原始连续色调照片的参考来源。此外,数据集还包括许多19世纪末和20世纪初出版的报纸复制品。这些复制品经常展示通过历史印刷和编辑过程引入的变换,包括:半色调印刷、裁剪、修图、覆盖绘画、对比度退化、污渍和数字化噪声。数据集的创建旨在支持历史图像流通、视觉相似性以及原始照片与其印刷复制品之间的跨域图像匹配研究。数据集结构分为两个主要文件夹:similar/ 包含已知复制品或同一原始照片的变体,并附有标注文件 groundtruth.xlsx;unique/ 包含被识别为独特实例的图像,没有在数据集中找到相应的复制品或重复项,并附有元数据文件 recto.csv。报纸来源包括《每日镜报》、《伦敦新闻画报》等多种历史报纸。潜在应用包括图像检索、视觉相似性学习、聚类、数字人文研究、文化遗产AI应用以及在严重印刷退化下的鲁棒性评估。
Halftone Revolution is a historical image dataset designed for image retrieval, similarity detection, and visual clustering research on archival newspaper imagery. The dataset contains digitized historical photographs originating primarily from the Victor Forbin collection and the Rol agency, which serve as reference sources for the original continuous-tone photographs. In addition, the dataset includes numerous newspaper reproductions published in the late 19th and early 20th centuries. These reproductions frequently exhibit transformations introduced through historical printing and editorial processes, including: halftone print, cropping, retouching, overpainting, contrast degradation, stains, and digitization noise. The dataset was created to support research on historical image circulation, visual similarity, and cross-domain image matching between original photographs and their printed reproductions. The dataset is divided into two main folders: similar/ contains images that correspond to known reproductions or variants of the same original photograph, with an annotation file groundtruth.xlsx; unique/ contains images identified as unique instances for which no corresponding reproduction or duplicate was found within the dataset, accompanied by the metadata file recto.csv. Newspaper sources include publications such as The Daily Mirror, The Illustrated London News, and others. Potential use cases include image retrieval, visual similarity learning, clustering, digital humanities research, cultural heritage AI applications, and robustness evaluation under severe print degradation.




