Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1999-07-29
2002-01-01
Couso, Yon J. (Department: 2623)
Image analysis
Pattern recognition
Feature extraction
C382S310000
Reexamination Certificate
active
06335986
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an apparatus and method for recognizing a pattern, and realizes to recognize characters, graphics, and symbols correctly depending on various states of input images when used with a printed character recognizing apparatus and a graphics recognizing apparatus as well as a handwritten character recognizing apparatus.
2. Description of the Related Art
Conventional handwritten character recognizing apparatuses such as an optical character reader (OCR) are designed to automatically reading characters written on an accounting list, etc. and automatically inputting the characters to eliminate the necessity of manually finding characters written on the accounting list, etc. and inputting the characters through a keyboard.
FIG. 1
is a block diagram showing the configuration of the conventional handwritten character recognizing apparatus.
In
FIG. 1
, a form/document
311
is read using a scanner to obtain a multiple-value image of the form/document
311
.
A preprocessing unit
312
binarizes a multiple-value image, removes noises, and amends the position of the form/document
311
.
Then, a character detecting unit
313
detects each character according to information about preliminarily defined ruled lines and positional information about a character.
A character recognizing unit
314
recognizes each character and outputs a character code. The character is recognized by collating each feature of an unknown character pattern detected by the character detecting unit
313
with the feature of each character category preliminarily entered in a recognizing dictionary
315
.
For example, a distance between feature vectors in a feature space is computed by converting a 2-dimensional character pattern into a feature vector in a feature space representing the feature of the character, as a similarity between the unknown character pattern and the character category preliminarily entered in the recognizing dictionary
315
. When the shortest distance is obtained between the feature vector of the unknown character pattern and the feature vector of the character category preliminarily entered in the recognizing dictionary
315
, the character category is recognized corresponding to the unknown character pattern.
A threshold is set for a distance between two feature vectors to avoid mistakenly recognizing a non-character such as a deletion line, a noise, a symbol, etc. for a character and outputting a character code for a non-character. If the distance between the two feature vectors is larger than the threshold, a reject code is output by determining that the unknown character pattern has no corresponding character category preliminarily entered in the recognizing dictionary
315
, or that the unknown character pattern refers to a non-character.
The recognizing dictionary
315
also contains the features of the character categories of high-quality characters, obscure characters, and deformed characters. A high-quality character recognizing dictionary
315
is referred to for high quality characters. An obscure character recognizing dictionary
315
is referred to for obscure characters. A deformed-character recognizing dictionary
315
is referred to for deformed characters. Thus, the difference in quality of the characters in the form/document
311
can be processed correspondingly.
FIG. 2
shows the configuration of the character recognizing apparatus for recognizing a character with a deletion line.
The character recognizing apparatus shown in
FIG. 2
comprises an image input unit
491
for inputting an original image containing a character and detecting or preprocessing a character from the input image, and an identifying unit
492
for identifying a character by extracting the feature of the character and comparing the extracted feature with the feature of the standard pattern stored in the recognizing dictionary.
When a character mistakenly entered in a form is removed with a deletion line, for example, six or more horizontal lines are entered on the character. It is determined that the character provided with six or more horizontal lines cannot be identified, and the character is rejected by the identifying unit
492
because it does not match any standard pattern stored in the recognizing dictionary.
However, the handwritten character recognizing apparatus shown in
FIG. 1
equally processes a detected character among obscure characters, deformed characters, high-quality characters using the same recognizing dictionary
315
.
Accordingly, there has been a problem that information about an obscure character entered in the recognizing dictionary
315
has a bad influence on the high-quality character recognizing process, and the obscure character entered in the recognizing dictionary
315
prevents high quality characters from being successful read.
In addition to obscure and deformed states, there are various environments for characters. For example, a character may touch its character box. When a single recognizing dictionary
315
is referred to in various environments, they affect each other, thereby generating a problem that the recognizing process cannot be performed with enhanced precision.
When the character recognizing apparatus shown in
FIG. 2
recognizes a character, six or more horizontal lines are required to delete an entered character using a deletion line. This is a heavy load to a user and therefore cannot be completely observed. As a result, a character with an apparent deletion line makes a small distance from a standard pattern stored in the recognizing dictionary and fails to be clearly distinguished from a character without a deletion line. Thus, the character to be deleted cannot be rejected and mistakenly read.
For example, as indicated by (A) shown in
FIG. 3
, the ‘0’ to be deleted is not rejected but recognized as ‘8’. As indicated by (B) shown in
FIG. 3
, the ‘1’ to be deleted is not rejected but recognized as ‘8’. As indicated by (C) shown in
FIG. 3
, the ‘7’ to be deleted is not rejected but recognized as ‘4’. As indicated by (D) shown in
FIG. 3
, the ‘6’ to be deleted is not rejected but recognized as ‘6’.
SUMMARY OF THE INVENTION
The present invention aims at providing a pattern recognizing apparatus and method capable of appropriately recognizing a character with high precision depending on the environment of the character.
According to the feature of the present invention, an input pattern is recognized by extracting the first predetermined feature from the input pattern and extracting the second predetermined feature from the input pattern from which the first feature has been extracted.
As a result, a recognizing process can be performed depending on each environment of a character.
According to other features of the present invention, a pattern is recognized by extracting the state of a process object from an input image and selecting a recognizing process suitable for the state for each process object.
Thus, a pattern recognizing process can be performed appropriately for each state on the input image having various states, thereby realizing the recognizing process with high precision.
According to other feature of the present invention, a state of a process object is extracted from an input image, and a pattern recognizing process exclusively for the first state is performed on the process object in the first state, and a pattern recognizing process exclusively for the second state is performed on the process object in the second state.
Thus, the recognizing process on the process object in the first state interact with the recognizing process on the process object in the second state, thereby successfully performing the recognizing processes with high precision.
According to other feature of the present invention, recognizing dictionaries are appropriately selected for an input image in various states.
For example, even if obscure characters, deformed characters, and high-quality characters are mixed in the input image, the recognizing process can be performed
Hotta Yoshinobu
Naoi Satoshi
Suwa Misako
Couso Yon J.
Fujitsu Limited
Staas & Halsey , LLP
LandOfFree
Pattern recognizing 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 Pattern recognizing apparatus and method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Pattern recognizing apparatus and method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2827634