您当前的位置: > 详细浏览

数字报版面布局自动生成方法

Automatic Generation Method of Digital Newspaper Layout

摘要:报纸版面对新闻有一个价值排序合理且美观新颖的展示,可以让读者面对众多新闻,在短时间获取最具价值的讯息和浏览乐趣。这是新闻报纸在众多媒体中的特色。然而,对于排版人员而言,手动制作直观、易读、美观的报纸版面布局往往需要耗费大量的时间成本。本文结合贝叶斯网络推断和约束规划技术,提出一种数字报版面布局自动生成方法。该方法首先基于历史版面数据驱动和专家经验对数字报版面的结构和属性建立推断模型,使得新生成的版面具有历史特定风格;然后利用推断结果建立混合整数约束规划模型计算版面布局,从而显著减少模型求解空间,提高布局质量。此外,推断模型提供多种可用候选结构为生成结果提供多样性,规划模型保证报纸版面内新闻不重叠、不溢出并具有良好的对齐性能。为了训练和验证模型,本文构建并公开了一个中文版面数据集。该数据集由数字报版面图片和相应的新闻内容组成,并带有详细版面新闻属性标记。最后,进行用户研究,结果表明了版面布局自动生成方法的有效性。

英文摘要:Newspaper pages have a reasonable and beautiful arrangement of news, which allows readers to quickly obtain valuable information and gain pleasure.This is the feature of newspaper in the mass media. However, for typesetters, generating an intuitive, readable and beautiful newspaper layout is a time-consuming manual task. Combining Bayesian network and constrained programming technology, this paper proposed an automatic generation method of digital newspaper layout. Firstly, based on the historical layout and expert experience, we learned and inferred the structure and key attributes of the digital newspaper layout, so that the newly generated layout has a history style; Then a mixed integer constrained programming model was proposed by the inference results, which significantly reduces the solution space of the model and improves the layout quality. Our inference model provides a variety of available candidate structures and the programming model ensures that the news in the newspaper layout does not overlap, does not overflow, and has good alignment performance. To train and validate the model, We construct and expose a Chinese newspaper layout data set. The data set is composed of digital newspaper page pictures and corresponding news content, with detailed news attribute tags. Finally, a user study is carried out, and the results show the effectiveness of our method.

版本历史

[V2] 2022-07-13 10:15:56 chinaXiv:202207.00112V2 下载全文
[V1] 2022-07-12 19:29:37 chinaXiv:202207.00112v1 查看此版本 下载全文
点击下载全文
同行评议状态
待评议
许可声明
metrics指标
  • 点击量1218
  • 下载量124
评论
分享
邀请专家评阅