Image analysis – Applications – Mail processing
Reexamination Certificate
1999-01-13
2003-03-18
Johns, Andrew W. (Department: 2721)
Image analysis
Applications
Mail processing
C382S179000, C382S187000
Reexamination Certificate
active
06535619
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an address recognition apparatus, and more specifically to an apparatus for reading an address from the handwritten characters in a free-pitch area.
2. Description of the Related Art
The conventional optical character recognition apparatus (OCR) specifies a place-name area, and reads an address by extracting a segmentation character (hereinafter referred to as a key character) which can be a delimiter of a place name as a result of recognizing characters one by one.
FIG. 1
is a block diagram showing the configuration of the conventional address recognition apparatus.
In
FIG. 1
, a character segmentation unit
501
segments a character for all possible segmentation candidates. A character recognition unit
502
outputs the recognition results of the first candidate through the Nth candidates in all possible candidates which can be segmented as a character. If any of the key characters ‘
’ (capital city), ‘
’ (prefecture), ‘
’ (prefecture), ‘
’ (prefecture), ‘
’ (city), ‘
’ (county), ‘
’ (ward), ‘
’ (town), ‘
’ (village), ‘
’ (area), etc. is included in the first through the Nth candidates, then a key character extraction unit
503
extracts the corresponding candidate as a key character.
A place-name area candidate retrieval unit
504
retrieves an area between key characters as a place-name area candidate. A place-name retrieval unit
506
compares character by character the first candidate through the Mth (M≦N) candidates with a place-name dictionary
505
, and retrieves a place-name in the place-name dictionary
505
as a place-name candidate if any one character of the place-name matches the first through Mth candidates. The place-name candidate evaluation value operation unit
507
computes the evaluation value of the a place-name candidate to obtain the most probable place-name candidate from among the place-name candidates retrieved by the place-name retrieval unit
506
. The place-name candidate selection unit
509
checks whether or not the place-name candidate retrieved by the place-name retrieval unit
506
is consistent with the place-name candidates before and after the present candidate. If it is consistent, the present place-name candidate is output as an address.
The conventional address recognition apparatus is described in, for example, the Tokukaihei 7-262320.
However, the conventional address recognition apparatus has the problem that it often makes a read error when it reads an address from a handwritten character string in a free-pitch area, and when characters to be read contact with each other. The. contact characters are read as one character.
There is also the problem with the conventional apparatus that the number of patterns to be recognized increases because all possible candidates segmented from a free-pitch area are recognized, thereby requiring a large amount of processes.
Furthermore, when any key character is contained in the first candidate through the Nth candidates, all the candidates are extracted as key characters. Thus, a non-key character is extracted as a key character, and a number of combinations of key characters for segmenting place-name areas appear, thereby requiring a large amount of processes.
If there are place-names, in the first through the Mth candidates, containing a character matching the character contained in the place-name entered in the place-name dictionary
505
, the place-names are all retrieved as place-name candidates. Therefore, a large number of place-name candidates are to be recognized, thereby requiring a large amount of processes to specify the address.
SUMMARY OF THE INVENTION
The first object of the present invention is to provide an address recognition apparatus capable of recognizing an address with high precision even if characters contact with each other.
The second object of the present invention is to provide an address recognition apparatus capable of recognizing an address with high precision.
To solve the above listed problems, the present invention includes a key character extraction unit for extracting a key character based on the result of separating contact characters; a place-name area extraction unit for extracting a place-name area based on the position of the key character; and a place-name recognition unit for recognizing the place-name of the place-name area based on the state of the contact characters before separated.
Even if characters forming a character string indicating an address contact with each other, the key character in the character string can be correctly extracted, the entire character string indicating a place-name can be extracted and processed in recognizing the place-name. Therefore, the process of segmenting a character string indicating a place-name into characters can be omitted, and the address can be efficiently recognized. Since it is not necessary to segment a character string indicating a place-name, mis-segmentation of the character string indicating the place-name can be avoided, thereby improving the correctness in address recognition.
According to an aspect of the present invention, the feature vector of the entire pattern segmented by a key character is compared with the feature vector of the place-name entered in the place-name entry dictionary so that the place-name can be recognized.
Thus, the character string indicating the place-name can be recognized without segmentation into single characters, thereby improving the efficiency and correctness in address recognition.
According to another aspect of the present invention, a place-name is entered for each attribute specified by a key character, and the feature vector of a pattern segmented by a key character is compared with a place-name having the attribute specified by a key character.
Thus, for example, if the attribute of the pattern segmented by a key character is ‘
’ (prefecture), a comparing operation is performed using a dictionary containing place-names of ‘
’ (prefecture), thereby performing the comparing operation corresponding to the attribute specified by a key character with improved recognition precision.
According to a further aspect of the present invention, when a connected pattern extracted from an input pattern is separated, the separated position is evaluated based on the size of an input pattern.
Thus, when a key character is extracted from an input pattern, the connected pattern can be separated for the size appropriate for extracting the key character. Therefore, the number of separation positions of the connected pattern can be reduced, and the number of times of the recognizing operations can also be reduced, thereby efficiently recognizing the address.
According to a further aspect of the present invention, a connected pattern for which a separation position can be detected based on the size of the connected pattern can be selected.
Thus, the separation position is detected for only a relatively large connected pattern which probably contains contact characters. For a small connected pattern not regarded as containing contact characters, the detection of a separation position can be omitted, thereby improving the efficiency in address recognition.
According to a further aspect of the present invention, if the value of the minimum point of the histogram of the number of black picture elements in the input pattern is equal to or smaller than a predetermined value, it is defined as a separation point candidate.
As a result, a narrow portion of a character can be distinguished from a contact point between characters, and only the contact point between characters can be detected with high precision. Thus, a connected pattern can be separated at a contact point with high precision.
According to a further aspect of the present invention, if the height-to-width ratio of a separated pattern refers to an area out of a predetermined range, the separation point is removed from separation point candidates.
Thus, the connected pattern can be prevented from being separated into portions of size inappr
Naoi Satoshi
Suwa Misako
Azarian Seyed
Fujitsu Limited
Johns Andrew W.
Staas & Halsey , LLP
LandOfFree
Address recognition apparatus and method does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Address recognition apparatus and method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Address recognition apparatus and method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3081918