Recognizing the Intended Message of Line Graphs: Methodology and Applications

Information graphics (line graphs, bar charts, etc.) are common in popular media and periodicals. They are usually included in such documents to convey a message. This dissertation discusses the processing of one kind of information graphic, namely a line graph. It presents a learned model for segmenting a line graph into visually distinguishable trends and a Bayesian network inference model that hypothesizes the intended message of the graph based on communicative signals in the graphic. Besides recognizing the intended message of line graphs, this dissertation also presents a method for identifying the paragraph in the document that is most relevant to its information graphic. The research results provided by this dissertation can be used for several purposes: to give blind individuals access to information graphics in an article, to provide the basis for a longer summary of the graphic, to build a summary that captures both the article and its containing information graphics, and to indicate a graphic’s content when indexing it for retrieval in a digital library.