Intelligent Text Extraction from PDF Documents

In recent years, PDF has become the de-facto standard for the exchange of print-oriented documents on the Web. This includes many business documents such as financial reports, newsletters and patent applications, and there are many commercial applications that require data to be extracted from these documents and processed by computer systems. A number of products currently exist on the market that navigate, extract and transform data from HTML pages; a process known as wrapping. One such methodology is Lixto, a product of research at our institute. However, none of these products are currently able to work with PDF files. We are investigating this possibility as part of the NEX-TWRAP project. This paper describes our work in progress, and details some of the low-level page segmentation techniques that we have investigated