项目名称: 面向移动阅读的复杂文档图像理解方法研究
项目编号: No.61300061
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 王勇涛
作者单位: 北京大学
项目金额: 23万元
中文摘要: 如何自动地将漫画书、文娱和体育类报刊等复杂版面出版物制作成适合于移动阅读的数字内容,是目前移动阅读发展所面临的瓶颈问题。复杂文档图像理解的目的,是实现这类出版物页面图像各构成对象的自动提取以及它们的阅读先后顺序的自动辨识,从而解决该瓶颈问题。现有的文档图像理解方法通常针对以文字为主体的文档图像,孤立地使用某个图像分析处理算法,局限性较大,无法处理这类包含大量的图形图像而且排版布局相对复杂的文档图像。本项目拟借鉴当前自然图像理解方法,使用能量最小化模型,研究一种更为通用有效的复杂文档图像理解方法。具体地,本项目将通过设计新的能量最小化函数及相应的优化算法,充分地使用相关先验知识,完成复杂文档图像理解中的多种构成对象提取任务以及不同构成对象联合识别任务。本项目研究成果将弥补现有文档图像理解方法的缺陷,为移动阅读内容制作提供关键技术支持,促进国内外移动阅读发展,因此本项目具有十分重要的研究意义。
中文关键词: 复杂文档图像理解;移动阅读;漫画分镜分割;立体几何对象三维重建;
英文摘要: How to automatically convert entertaining publications such as comic books and sports magazines into digital contents that are suitable to display on mobile devices is the bottleneck problem of mobile reading. Complex document image understanding aims to solve this problem by automatically detecting each object that composes the whole image page and then indentifying their reading orders. The existing document image understanding methods are specifically designed to process the document images mainly composed of texts, and solely exploit certain image processing algorithm, thus can't handle such kind of complex document image which mainly consists of graphics with complex layout. This project aims to propose a new more general and efficient document image understanding method by using the methodology of the state-of-the-art natural image understanding and the energy minimization method. In detail, the proposed method shall conduct the tasks of object detection and joint recognition for multiple detected objects by developing the new energy minimization function and the corresponding optimization algorithms. It is expected to overcome the drawbacks of the existing document understanding method, and provide key technical support for producing mobile reading contents, thus promote the development of both domestic a
英文关键词: complex document image understanding;mobile reading;comic panel detection;solid geometric object reconstruction;