INFORMys: a flexible invoice-like form-reader system

Abstract

In this paper, we describe a flexible form-reader system capable of extracting textual information from accounting documents, like invoices and bills of service companies. In this kind of document, the extraction of some information fields cannot take place without having detected the corresponding instruction fields, which are only constrained to range in given domains. We propose modeling the document's layout by means of attributed relational graphs, which turn out to be very effective for form registration, as well as for performing a focussed search for instruction fields. This search is carried out by means of a hybrid model, where proper algorithms, based on morphological operations and connected components, are integrated with connectionist models. Experimental results are given in order to assess the actual performance of the system.

Semantic information extraction from images of complex documents

“…Cesarini et al [6] have developed a system with similar goals to those of our proposed solution. We have improved upon their work by focusing on usability and simplifying the work of the human user in charge of building a conceptual model for a document, which then is used as the basis for matching, interpreting and extracting contents from document images.…”

“…Even though we find high-quality results in all systems referred to in the previous paragraphs, many of them have employed techniques which are incompatible either with the kind of document with which we are dealing or with the requirements we have established for our work: -In some of these works [6,9] graphical elements such as logos and, especially, horizontal and vertical lines are used to model the structure of the input document. However, when dealing with documents which feature complex background patterns, logos and lines can be very difficult to detect.…”

“…Following Cesarini et al [6,7], we have represented document models and instances using Attributed Relational Graphs in which the main attribute of each edge is the relative position between the connected elements. Following Duygulu and Atalay [9], we have adopted in our document model a hierarchical structure, since we allow text fields to depend on the position of other text fields.…”

Adaptive and Interactive Approaches to Document Analysis

“…This approach is particularly suitable for limited-vocabulary applications like postal addresses or legal amounts on bank checks [35,36], and for correction of OCR errors [37]. Another example of holistic word classification is based on statistics extracted from each cell of a grid superimposed on the word [38]. For degraded documents with larger vocabularies, word level indexing (as opposed to keyword spotting) was proposed with a three-stage comparison based on word aspect ratios, vector features extracted with a grid superimposed on each word, and within-word connectivity.…”

Thick 2D relations for document understanding

“…More abstract representations are labeled and weighted graphs. These have been used in various systems such as, for instance, the ones presented in (Li and Ng, 1999;Cesarini et al, 1998;Walischewski, 1997). …”