Image analysis – Image transformation or preprocessing – Image storage or retrieval
Reexamination Certificate
1999-09-02
2003-09-16
Mehta, Bhavesh M. (Department: 2625)
Image analysis
Image transformation or preprocessing
Image storage or retrieval
C382S181000, C707S793000
Reexamination Certificate
active
06621941
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to pattern localization and, more particularly, to a system combining two dimensional pattern localization with text recognition to enable indexing text keyword extraction.
BACKGROUND OF THE INVENTION
Image content-based retrieval is becoming a powerful alternative/addition to conventional text annotation-based retrieval. Even so, it has yet to reach the robustness and computational effectiveness of text-based retrieval. Text-based retrieval, on the other hand, is notoriously lacking in precision, even when boolean combinations of key-words are allowed. It is a common observation with those using popular conventional search that full text indexing of documents (scanned or electronic) causes a large number of irrelevant documents to be retrieved.
A more productive use of text-based querying is when it is combined with image content-based querying. A special case of this occurs when the text strings relevant for indexing documents occur within image structures, such as text in special regions of a news video or text within region fields of a form. Retrieval based on such structured text can yield fewer but more relevant matching documents.
An example of the above-mentioned special case arises in the area of processing engineering drawing documents, a large number of which still exist in paper form. Creating electronic conversion of such documents is an important business for large format scanner makers. As is known, large format scanners can scan engineering drawing documents at a relatively fast rate of 25 sheets/minute, and are quickly giving rise to very large databases (in excess of 100,000 objects) of large-sized drawing images (e.g., 14000×9000 pixels). Currently, indexing of such documents is done manually with skilled keyboard operators, and is considered a highly labor intensive activity constituting a significant cost in the digitizing of scanned images. Manual indexing by a keyboard operator can also be unreliable since the keywords employed by a user may not match the ones attached to the documents during database creation.
In contrast to full-text indexing of pure text documents, automatic full-text indexing using conventional OCR algorithms will not yield useful results for drawing images. Fortunately, useful text information for indexing such drawing images is found in specific image structures called “title blocks”. Typically, a title block will include information pertinent for indexing a corresponding drawing, such as part number, name of the unit being depicted, date of design, and architect name. Indexing keyword extraction from such image structures requires that the image structures themselves be first identified.
As will appear from the Detailed Description below, the present invention employs some of the principles underlying a solution for a model indexing problem, namely the principles underlying “Geometric Hashing”. Referring to articles by Y. Lamdan and H. J. Wolfson (entitled “Geometric hashing: A general and efficient model-based recognition scheme”, in Proceeding of the International Conference on Computer Vision, pages 238-249, 1988, and “Transformation invariant indexing” in Geometric Invariants in Computer Vision, IT Press, pages 334-352, 1992), Geometric Hashing has been used to identify objects in pre-segmented image regions. Another work extending the basic geometric hashing scheme for use with line features includes an article by F.C.D. Tsai entitled “Geometric hashing with line features” in Pattern Recognition, Vol. 27, No. 3, pages 377-389, 1994. An extensive analysis of the geometric hashing scheme is provided in an article by W. E. L. Grimson and D. Huttenlocher entitled “On the sensitivity of geometric hashing”, in Proceedings International Conference on Computer Vision, pages 334-339, 1990.
Obtaining suitable geometric hash functions has also been explored in an article by G. Bebis, M. Georgiopolous and N. Lobo entitled “Learning geometric hashing functions for model-based object recognition” in Proceedings International Conference on Computer Vision, pages 543-548, 1995, and a discussion of using the concept of “rehashing” in the context of geometric hashing is provided in an article by I. Rigoustos and R. Hummel “Massively parallel model matching: Geometric hashing on the connection machine” in IEEE Computer, pages 33-41, February 1992.
As mentioned above, manual indexing of drawing documents can be time-consuming and thus undesirable. Moreover, pertinent indexing text may be more readily obtainable from two dimensional patterns (e.g., title blocks) embedded in previously stored drawing documents. Identifying these embedded patterns, however, can be difficult since orientation and/or location of the patterns, relative to many other adjacent patterns, is typically unknown. It would be desirable to provide an indexing system that obtains indexing information automatically from a previously stored first set of drawing documents by localizing a query image with respect to the previously stored first set of drawing documents and then extracts corresponding text therefrom for use in storing a second set of drawing documents.
SUMMARY OF THE INVENTION
In accordance with the present invention there is provided a document processing system, comprising: a memory; a plurality of model images stored in said memory, the model images being represented by a first information set with the first information set varying as a function of object-based coordinates, at least one of the plurality of model images including a text containing region having index information intended for use in storing one or more document pages; a query image represented as a second set of information varying as a function of object-based coordinates; an image localization module, communicating with said memory, for corresponding the second set of information with a portion of the first set of information to obtain the text containing region; and a text extraction module, communicating with said image localization module, for extracting the index information from the text containing region to facilitate the storing of the one or more document pages.
REFERENCES:
patent: 5742807 (1998-04-01), Masinter
patent: 5897637 (1999-04-01), Guha
patent: 5953451 (1999-09-01), Syeda-Mahmood
patent: 6134338 (2000-10-01), Solberg et al.
Tanveer Syeda-Mahmood (Indexing of Handwritten Document Images)--1997 IEEE.*
Yehezkel Lamdan and Haim J. Wolfson, “Geometric Hashing: A general and Efficient Model-Based Recognition Scheme”, 1/88, pp. 238-249.
Haim J. Wolfson and Yehezkel Lamdan, “Transformation Invariant Indexing”, in Geometric Invariants in Computer Vision, IT Press, pp. 334-352, 1992.
Frank C.D. Tsai, “Geometric Hashing with Line Features”, Pattern Recognition, vol. 27, No. 3, pp. 377-389, 1994.
W. Eric L. Grimson and Daniel P. Huttenlocher, “On Sensitivity of Geometric Hashing”, in Proceedings International Conference on Computer Vision, pp. 334-339, 1990.
George Bebis, Michael Georgiopoulos and Niels Da Vitroia Lobo, “Learning Geometric Hashing Functions for Model-Based Object Recognition”, in Proceedings International Conference on Computer Vision, pp. 334-339, 1990.
Isidore Rigoutsos and Robert Hummel, “Massively Parallel Model Matching: Geometric hashing on the connection machine”, in IEEE Computer, pp. 33-41, Feb. 1992.
Handley John C.
Syeda-Mahmood Tanveer F.
Blair Philip E.
Cohen Gary B.
Mehta Bhavesh M.
Patel Kanji
Xerox Corporation
LandOfFree
System of indexing a two dimensional pattern in a document... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System of indexing a two dimensional pattern in a document..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System of indexing a two dimensional pattern in a document... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3058959