Automatic Processing, Classification and Retrieval of Unconstrained Digital Documents

在许多组织中，出于存档目的，必须由人类处理的纸质文档的数量是巨大的，并且每天都在增长。迫切需要实现这一过程的自动化。对于干净的印刷文件，这个问题至少在理论上可以认为已经得到了解决。然而，困难在于处理不受限制的手写文档。据我们所知，到目前为止，还没有能够基于任意文本关键字对文档图像进行分类的系统。最近，在邮件分拣方面取得了一些成功。然而，在邮件分拣中，邮政编码是以某种格式编写的。无约束数字文档中的关键词检测是文档分类和检索应用中的一个具有挑战性的问题，因此近年来引起了学术界和工业界的极大兴趣。关键字检测在不受约束的文档中实现两个目的。它通过计算机程序确定扫描的文档图像是否包含文本关键字，并且可选地在文档图像中点出该关键字的实例。我们项目的目标是开发有效的技术来处理和分类不受限制的数字文档。我们希望开发一个通用的关键字检测系统，它将允许用户对输入的文档图像进行分类，并检索包含一些必要信息的文档。

The amount of paper documents that must be processed by humans in many organizations both commercial and government offices for archival purposes is huge and growing every day. There is an urgent need for automation of this process. For clean printed documents, the problem could be considered as already solved at least in theory. However, the difficulty is in dealing with unconstrained handwritten documents. To the best of our knowledge, to date there is no system capable of classifying document images based on arbitrary text keywords. Recently there has been some success in mail sorting. However in mail sorting the postal codes are written in certain format. The detection of keywords in unconstrained digital documents is a challenging problem in document classification and retrieval applications, and thus has attracted considerable interest by academia and industry in recent years. Keyword detection fulfills two purposes in unconstrained documents. It determines by means of a computer program whether or not a scanned document image contains a text keyword and optionally, spots the instances of the keyword in the document image. The goal of our project is to develop efficient techniques for processing and classification of unconstrained digital documents. We wish to develop a general keyword detection system which will allow the user to classify input document images and to retrieve the documents which contain some necessary information.