Page Similarity and Classification

Abstract: Document analysis and recognition techniques address several types of documents ranging from small pieces of information such as forms to larger items such as maps. In most cases, humans are capable of discerning the type of document and therefore its function without reading the actual textual content. This is possible because the layout of one document often reflects its type. For instance, invoices are more visually similar to one another than they are to technical papers and vice versa. Two related tasks, page classification and page retrieval, are based on the analysis of the visual similarity between documents and are addressed in this chapter. These tasks are analyzed in this chapter in a unified perspective because they share several technical features and are sometimes adopted in common applications.