Structural extraction from visual layout of documents

Abstract

Most information extraction systems focus on the textual content of the documents. They treat documents as sequences of words, disregarding the physical and typographical layout of the information. While this strategy helps in focusing the extraction process on the key semantic content of the document, much valuable information can also be derived form the document physical appearance. Often, fonts, physical positioning and other graphical characteristics are used to provide additional context to the information. This information is lost with pure-text analysis.

Web news categorization using a cross-media document graph

“…al. [19] are based on templates that characterize each part of the document. These templates are either extracted manually or semi-automatically.…”

“…al. [19] devised a learning algorithm to extract information (author, title, date, etc) that relies on a general procedure for structural extraction. Their proposed technique enables the automatic extraction of entities from the document based on their visual characteristics and relative position in the document layout.…”

“…Zhuang [25] present a method for cross-media retrieval, in which cross-media features are integrated with multimedia data via a cross-reference graph model so as to improve retrieval accuracy progressively by learning associations between objects present in the model. We note that these two works have improved on the work in [19] by identifying the relations between different media objects . But whereas these approaches need to be trained using supervised structure learning algorithms, our canonical representation of documents is completely unsupervised and generic enough to support a wide range of multimedia data types.…”

Applying semantic web technologies to knowledge sharing in aerospace engineering

“…The document layout and extracted cross-references (e.g., captions) may suggest how each text segment relates to each image, examples include (Arasu and Garcia-Molina 2003;Crescenzi et al 2001;Rosenfeld et al 2002). Arasu and Garcia-Molina (2003), Crescenzi et al (2001) and Rosenfeld et al (2002) approaches are based on (manually or semi-automatically extracted) templates that characterise each part of the document. Rosenfeld et al (2002) implement a learning algorithm to extract information such as the author, title and date.…”

Structure Extraction from Presentation Slide Information

“…Rosenfeld et al [6] and Zhai et al [9] suggested a structure extraction method for PDF and Web documents using probabilistic approaches, such as the machine-learning and tree-graph-matching algorithms, respectively. These approaches need to prepare a large amount of annotated data, and the models made from the data are dependent on the data.…”