AREA EXTRACTION PROGRAM, CHARACTER RECOGNITION PROGRAM, AND CHARACTER RECOGNITION DEVICE

• US 20090202152A1
• Filed: 02/05/2009
• Published: 08/13/2009
• Est. Priority Date: 02/12/2008
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Fujimoto, Katsuhito, TAKEBE, Hiroaki
• Granted Patent: US 8,300,942 B2
• Time in Patent Office: Days
• US Class Current: 382/185
• CPC Class Codes: G06F 18/29 Graphical models, e.g. Bayesian networks
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/158 using character size, text spacings or pitch estimation
• CPC Class Codes: G06V 30/19187 Graphical models, e.g. Bayesian networks or Markov models
• CPC Class Codes: G06V 30/262 using context analysis, e.g. lexical, syntactic or semantic context

Abstract

An area extraction method including obtaining a character lattice showing a connection relation between unit areas, which are obtained by separating a character string pattern in an image into patterns each recognized as corresponding to a single character, judging whether or not all combinations of each of the unit areas in the obtained character lattice and each of the unit areas in a regular lattice defining a regular connection relation between the unit areas are likely to be established, generating a path coupling between nodes corresponding to the combination of the unit areas which is determined as likely to be established, determining an optimum path from the generated paths based on a degree of coincidence with the regular lattice or the character lattice, and extracting from an image the unit areas in the character lattice corresponding to the determined optimum path.

Claims

1. A computer readable recording medium in which a program executing an area extraction method is recorded, the program when executed by a computer causes the computer to perform the method comprising:

an obtaining step of obtaining a character lattice showing a connection relation between unit areas, which are obtained by separating a character string pattern in an image into patterns each recognized as corresponding to a single character;

a judging step of judging whether or not all combinations of each of the unit areas in the character lattice obtained in the obtaining step and each of the unit areas in a regular lattice defining a regular connection relation between the unit areas are likely to be established;

a generating step of generating a path coupling between nodes corresponding to the combination of the unit areas, which is determined as likely to be established in the judging step;

a determining step of determining an optimum path from the paths generated in the generating step based on a degree of coincidence with the regular lattice or the character lattice; and

an extracting step of extracting, from an image, the unit areas in the character lattice corresponding to the optimum path determined in the determining step.

2. The area extraction program as claimed in claim 1, wherein the generating step further comprises:

a node extracting step of extracting a newly coupling node newly coupled to an other node and a preceding node which has been already coupled to other node and is in the combination of the unit areas in a predetermined range from the unit area corresponding to the newly coupling node in the connection relation between the unit areas shown by the character lattice and the regular lattice; and

a node coupling step of coupling the preceding node, which is extracted in the node extracting step and has the largest number of coupling to the preceding node, to the newly coupling node.

3. The area extraction program as claimed in claim 2, wherein, in the coupling step, when there are a plurality of preceding nodes with the largest coupling number, regarding the corresponding plurality of preceding nodes, the preceding node included in the path and the newly coupling node are coupled to each other based on a reproducing ratio of reproducing the connection relation between the unit areas shown by the regular lattice.

4. The area extraction program as claimed in claim 1, wherein, in the determining step, the optimum path is determined based on a reproducing ratio of reproducing the connection relation between the unit areas shown by a regular lattice.

5. The area extraction program as claimed in claim 4, wherein, in the determining step, when there are a plurality of paths with the largest reproducing ratio, regarding the corresponding plurality of paths, the optimum path is determined based on a matching ratio of fitting to the connection relation between the unit areas shown by the character lattice.

6. The area extraction program as claimed in claim 1, wherein the judging step further comprises:

a calculating step of, with respect to patterns in the unit area in the character lattice in each of the combinations of the unit areas, calculating a recognition reliability showing a ratio between a character recognition accuracy in a case where character recognition is executed using a dictionary for characters including all characters and the character recognition accuracy in a case where the character recognition is executed using a dictionary for categories in which only characters in a category, to which the unit areas in the regular lattice belong, is included,

wherein a combination in which the recognition reliability calculated in the calculating step meets a predetermined standard is determined to be likely to be established.

7. A computer readable recording medium, in which a program executing an area extraction method is recorded, the program when executed by a computer causes the computer to perform the method comprising:

an obtaining step of obtaining a character lattice showing a connection relation between unit areas, which are obtained by separating a character string pattern in an image into patterns each recognized as corresponding to a single character;

a judging step of judging whether or not all combinations of each of the unit areas in the character lattice obtained in the obtaining step and each of the unit areas in a regular lattice defining a regular connection relation between the unit areas are likely to be established;

a generating step of generating a path coupling between nodes corresponding to the combination of the unit areas, which is determined as likely to be established in the judging step;

a determining step of determining an optimum path from the paths generated in the generating step based on a degree of coincidence with the regular lattice or the character lattice;

an extracting step of extracting, from an image, the unit areas in the character lattice corresponding to the optimum path determined in the determining step; and

a recognizing step of applying character recognition to a pattern in the unit area, extracted in the extracting step, with use of a dictionary for categories, including only characters in a category to which the unit areas in the regular lattice belong.

8. A character recognition device comprising:

acquisition means for acquiring a character lattice showing a connection relation between unit areas, which are obtained by separating a character string pattern in an image into patterns each recognized as corresponding to a single character;

judgment means for judging whether or not all combinations of each of the unit areas in the character lattice acquired by the acquisition means and each of the unit areas in a regular lattice defining a regular connection relation between the unit areas are likely to be established;

generating means for generating a path coupling between nodes corresponding to the combination of the unit areas, which is determined as likely to be established by the judging means;

determining means for determining an optimum path from the paths generated by the generating means based on a degree of coincidence with the regular lattice or the character lattice;

extraction means for extracting, from an image, the unit areas in the character lattice corresponding to the optimum path determined by the determining means; and

recognition means for applying character recognition to a pattern in the unit area, extracted by the extraction means, with use of a dictionary for categories, including only characters in a category to which the unit areas in the regular lattice belong.

9. A character recognition method comprising:

an obtaining step of obtaining a character lattice showing a connection relation between unit areas, which are obtained by separating a character string pattern in an image into patterns each recognized as corresponding to a single character;

a judging step of judging whether or not all combinations of each of the unit areas in the character lattice obtained in the obtaining step and each of the unit areas in a regular lattice defining a regular connection relation between the unit areas are likely to be established;

a generating step of generating a path coupling between nodes corresponding to the combination of the unit areas, which is determined as likely to be established in the judging step;

a determining step of determining an optimum path from the paths generated in the generating step based on a degree of coincidence with the regular lattice or the character lattice;

an extracting step of extracting from an image the unit areas in the character lattice corresponding to the optimum path determined in the determining step; and

a recognizing step of applying character recognition to a pattern in the unit area, extracted in the extracting step, with use of a dictionary for categories, including only characters in a category to which the unit areas in the regular lattice belong.

10. A character recognition device, comprising:

an acquisition part acquiring a character lattice showing a connection relation between unit areas, which are obtained by separating a character string pattern in an image into patterns each recognized as corresponding to a single character;

a judgment part judging whether or not all combinations of each of the unit areas in the character lattice acquired by the acquisition part and each of the unit areas in a regular lattice defining a regular connection relation between the unit areas are likely to be established;

a generating part generating a path coupling between nodes corresponding to the combination of the unit areas, which is determined as likely to be established by the judging part;

a determining part determining an optimum path from the paths generated by the generating part based on a degree of coincidence with the regular lattice or the character lattice;

an extraction part extracting, from an image, the unit areas in the character lattice corresponding to the optimum path determined by the determining part; and

a recognition part applying character recognition to a pattern in the unit area, extracted by the extraction part, with use of a dictionary for categories, including only characters in a category to which the unit areas in the regular lattice belong.