Image analysis – Image segmentation – Segmenting individual characters or words
Reexamination Certificate
1999-01-22
2001-03-06
Tran, Phuoc (Department: 2721)
Image analysis
Image segmentation
Segmenting individual characters or words
C382S203000, C382S204000
Reexamination Certificate
active
06198846
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to character recognition systems and, more particularly, to character recognition systems for automatically recognizing characters written on paper.
FIG. 18
shows a system for recognizing characters in contact with one another. In this system, contact portions of characters in a character pattern are detected, and the pattern is separated at the detected contact portions into character candidates, and kinds of the separated character candidates are determined in a character recognition mean.
Specifically, the illustrated character recognition system comprises a character row extracting means
101
for extracting an area of one character row, a character candidate separating means
201
for separating areas each corresponding to one character, a character recognition means
301
for recognizing each character candidate separated in the character candidate separating means
201
, and a recognition result verifying means
401
for verifying the result of recognition executed in the character recognition means
301
.
Japanese Patent Laid-Open No. 59-098283 discloses a pattern separating and recognizing system as a first example of the character recognition system having the above construction. In this system, character separation result data which are obtainable for, for instance, a contact character pattern as shown in
FIG. 19
, as outputs of the character candidate separating means
201
and include incorrect character data, are registered in a character recognition dictionary
302
. When a candidate character pattern hits the dictionary obtained from the incorrect character separation result, the character recognition means
301
provides data representing the kind of the hit character and also data representing that the hit character is obtained by separating the pattern at incorrect position, i.e., an incorrect partial pattern. The recognition result verifying means
401
provides recognition result data when and only when it verifies that both the results of recognition of the left and right separated character candidates are not partial patterns.
In
FIG. 19
, the three arrows show examples of the character candidate separating position. Assuming that the right end arrow represents the correct separating position, when the pattern separation is performed at this position, the means
401
verifies that both the recognition results of the left and right separated character candidates are not partial patterns (“0”), and provides data “00”.
Japanese Patent Laid-Open No. 8-16720 discloses a second example of the prior art character recognition system. In this system, the character candidate separating means
201
separates a contact character pattern at a minimal position, which is detected from a horizontal projection profile of a character row pattern. When no minimal position is detected from the profile, the contact character pattern is collectively recognized without separating it. To this end, the character recognition means
301
has a special structure that can collectively recognize, without separation, a pattern which is incapable of being separated by using the projection profile, and reference patterns for contact characters are stored in the character recognition dictionary
302
.
Japanese Patent Laid-Open No. 1-181176 discloses a third example of the prior art character recognition system. In this character recognition system, the character candidate separating means
201
includes means for detecting characters not in contact with other characters, and means for detecting the contact characters. In this latter system, unlike the preceding second example of system, the separating position is not absolutely determined by using a pure horizontal projection profile of the character row pattern. Instead, a contact character pattern is divided into two divisions for each pixel in the horizontal direction. A position is then detected, which provides the best mean evaluation value of the recognition results obtained, and the pattern is separated at this position.
In any of the above three prior art character recognition systems, a plurality of separated character candidates are prepared for each contact character pattern, and the best combination of these candidates is selected as a character recognition result.
However, as shown in
FIG. 20
, depending on the way of separated character candidate selection, it is possible that the character recognition means recognizes a wrong character as a result of pattern separation at a wrong separating position. In the case of
FIG. 20
, the character recognition means erroneously recognizes the right character candidate to be figure “2”.
As described above, the above prior art character recognition systems have a high possibility that the character recognition means erroneously recognizes a wrong character kind from a character candidate pattern obtained by separating a contact character pattern at a wrong separating position.
SUMMARY OF THE INVENTION
An object of the invention, accordingly, is to provide a character recognition system, which can reduce erroneous reading of the character recognition means, i.e., reduce the possibility that the character recognition means erroneously recognizes a wrong character kind from a character candidate pattern obtained by separating a contact character pattern at a wrong separating position, thus improving the reading performance.
According to an aspect of the present invention, there is provided a character recognition system comprising: a first means for extracting a character row containing contact characters, detecting character separating position candidates in the derived character row and separating areas each corresponding to one character as character candidate patterns from the derived character row; a second means for determining the separation shape of each separated character candidate pattern; and a third means for determining the character kind from each separated separated character candidate and the corresponding separation shape, making a check, on the basis of the character recognition result of each character candidate pattern and the corresponding separation shape, as to whether the combination of the character recognition result and the separation shape is right, and providing the recognition result if the combination is right.
According to another aspect of the present invention, there is provided a character recognition system for detecting contact positions in a contact character pattern containing contact characters, separating character candidates from the contact character pattern at the contact positions thereof as separating positions, and determining the character kind of each character candidate in a character recognition unit, wherein: the shapes of the separating positions are detected when separating the character candidates from the contact character pattern, a character recognition dictionary with reference patterns stored therein is used to provide a character kind and a separating position shape most resembling a reference pattern for each detected separating position shape, and for a combination of a character recognition result and the corresponding separation shape the recognition result is provided if the combination is right.
According to other aspect of the present invention, there is provided a character recognition system comprising: a character row extraction means for extracting a character row containing contact characters; a separating position candidate detecting means for detecting character separating position candidates in the character row derived by the character row deriving means; a character candidate separating means for separating areas each corresponding to one character as character candidate patterns by selecting the separating position candidates detected by the separating position candidate detecting means; a separation shape determining means for determining the separation shapes of the separating position candidates detected by the separating position candidate de
Foley & Lardner
NEC Corporation
Tran Phuoc
LandOfFree
Character recognition system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Character recognition system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Character recognition system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2457496