Method for graph-based table recognition

• US 5,841,900 A
• Filed: 01/11/1996
• Issued: 11/24/1998
• Est. Priority Date: 01/11/1996
• Status: Expired due to Term

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Rahgozar, M. Armon, Cooperman, Robert
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/585,142
• Time in Patent Office: 1,048 Days
• Field of Search: 382/202, 382/177, 382/176, 382/180, 382/306, 395/779, 395/770, 395/782, 395/774, 395/776, 707/517, 707/509, 707/520, 707/513, 707/514
• US Class Current: 382/176
• CPC Class Codes: G06V 30/416 Extracting the logical structure, e.g. chapters, sections or page numbers; Identifying elements of the document, e.g. authors

Abstract

The present invention is a method for bottom-up recognition of tables within a document. This method is based on the paradigm of graph-rewriting. First, the document image is transformed into a layout graph whose nodes and edges represent document entities and their interrelations respectively. This graph is subsequently rewritten using a set of rules designed based on apriori document knowledge and general formatting conventions. The resulting graph provides a logical view of the document content. It can be parsed to provide general format analysis information.

Claims

1. A graph-based method for recognizing tables present in a document represented as digital image data, including the steps of:

segmenting the digital image data to identify textual and image entities within the document;

building a layout graph of the document using the entities;

automatically tagging each of the text entities with a label from a document node alphabet to produce a labeled graph;

automatically rewriting the labeled graph, by manipulating the text entities therein, using at least one rewriting rule, to identify the logical structure of the document, wherein the logical structure includes an identified table and the labeled graph is a host graph and where the step of automatically rewriting the labeled graph by manipulating the text entities comprises the steps of

identifying at least two entities in the layout graph that are isomorphic instances of subgraphs, and

replacing, in the host graph, the at least two entities that are isomorphic instances of the subgraphs with a nonterminal entity.

2. The method of claim 1, wherein the segmentation operation identifies as text entities contiguous, non-overlapping regions of text.

3. The method of claim 2, wherein the contiguous, non-overlapping regions of text each have a bounding box identifying the boundaries thereof.

4. The method of claim 3, wherein the step of building a layout graph is based upon a set of geometric relations between entity bounding boxes.

5. The method of claim 4, wherein the geometric relations between entity bounding boxes are stored in a memory as an ordered list.

6. The method of claim 4, wherein the geometric relations between entity bounding boxes results in the construction of an edge whenever a relation therebetween is detected from the group consisting of:

left of;

right of;

top of; and

bottom of.

7. The method of claim 2, wherein the text entities are selected from at least one of:

a character;

a word;

a line; and

a text region.

8. The method of claim 7, wherein the text entity group obeys a hierarchical structure, where

a character is subordinate to a word;

a word is subordinate to a line; and

a line is subordinate to a text region.

9. The method of claim 7, further comprising the step of re-classifying text regions according to their logical identity as one of a plurality of classifications selected from at least one of:

a paragraph;

a column;

a table;

an indexed list;

a jagged text region; and

an unformatted text region.

10. The method of claim 1, wherein the step of manipulating the text entities of the labeled graph includes a context-free rewrite rule.

11. The method of claim 1, wherein the step of manipulating the text entities of the labeled graph includes a context-sensitive rewrite rule.

12. The method of claim 1, wherein the nonterminal entity is connected to the remainder of the host graph based upon a set of embedding relations.

13. The method of claim 12, wherein the embedding relations are geometric.

14. The method of claim 12, wherein the embedding relations are logical.

15. The method of claim 12, wherein each rewrite rule has at least one application condition associated therewith.

16. The method of claim 15, further comprising the step of testing the application condition prior to the application of a rewrite rule.

17. The method of claim 16 wherein the application condition requires that a bounding box about the nonterminal entity does not overlap with a bounding box of any of the other entities in the host graph.

18. The method of claim 1 wherein each rewrite rule has an attribute transfer function associated therewith.

19. The method of claim 1 wherein each rewrite rule has a production rule associated therewith.

20. The method of claim 1, further comprising the step of recursively applying the step of manipulating the text entities until no more rewrite rules are applicable.

21. A graph-based method for recognizing tables present in a document represented as digital image data, including the steps of:

segmenting the digital image data to identify textual and image entities within the document;

building a layout graph of the document using the entities;

automatically tagging each of the text entities with a label from a document node alphabet to produce a labeled graph;

automatically rewriting the labeled graph, by manipulating the text entities therein, using at least one rewriting rule, to identify the logical structure of the document, wherein the logical structure includes an identified table and where the rewriting rule comprises

a production rule,

an embedding function,

an attribute transfer functions, and

an application condition.

22. An automated document structure recognition system for performing a graph-based method for recognizing tables present in a document represented as digital image data, including:

a segmentation subsystem for segmenting the digital image data to identify textual and image entities within the document;

a graph construction module for building a layout graph of the document using the entities;

an entity recognition subsystem for automatically tagging each of the text entities within the layout graph with a label from a document node alphabet to produce a labeled graph; and

a graph rewriting system for automatically rewriting the labeled graph, by manipulating the text entities therein, using at least one rewriting rule, to identify the logical structure of the document, wherein the logical structure includes an identified table, wherein the labeled graph is a host graph and where the rewriting system identifies at least two entities in the layout graph that are isomorphic instances of subgraphs, and replaces, in the host graph, the at least two entities that are isomorphic instances of the subgraphs with a nonterminal entity.