Character recognition system

A character recognition system comprising:

• a computer that performs character recognition by comparing distances between a feature vector extracted from an input pattern and reference vectors of previously stored target characters, wherein all characters are classified into a plurality of classes based on occurrence frequencies of the target characters,
• a dictionary storage device for storing the reference vectors for each of the classes and a computer are included,
• wherein the computer is programmed to calculate distances between the feature vector and the reference vectors of a plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes, and to select, when there is a distance smaller than a class threshold determined for the class, a character of a reference vector corresponding to the distance as a character candidate, and
• a similar character dictionary for providing characters having similar relations to arbitrary characters, wherein the computer is programmed to take out characters similar to the respective character candidates from the similar character dictionary in response to a selection of the plurality of character candidates, to calculate distances between the feature vector to be recognized and reference vectors of the similar characters, to sort distance values of the plurality of character candidates and distance values of the similar characters, and to select a predetermined number of characters in ascending order of distance as recognition character candidates.

A character recognition system comprising:

• a computer that performs character recognition by comparing distances between a feature vector extracted from an input pattern and reference vectors of previously stored target characters, wherein all characters are classified into a plurality of classes based on occurrence frequencies of the target characters,
• a dictionary storage device for storing the reference vectors for each of the classes and a computer are included,
• wherein the computer is programmed to calculate distances between the feature vector and the reference vectors of a plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes, and to select, when there is a distance smaller than a class threshold determined for the class, a character of a reference vector corresponding to the distance as a character candidate, wherein the computer is programmed to calculate distances between the feature vector and the reference vectors of a plurality of characters belonging to a class having a second highest character occurrence frequency when there is no distance smaller than the class threshold, and to select, when there is a distance smaller than a second class threshold determined for the class, a character of a reference vector corresponding to the distance as a candidate character, and
• a similar character dictionary for providing characters having similar relations to arbitrary characters, wherein the computer is programmed to take out characters similar to the respective character candidates from the similar character dictionary in response to selection of the plurality of character candidates, to calculate distances between the feature vector to be recognized and reference vectors of the similar characters, to sort distance values of the plurality of character candidates and distance values of the similar characters, and to output a predetermined number of characters in ascending order of distance as recognition character candidates.

A character recognition system comprising:

• a computer that performs character recognition by comparing distances between a feature vector extracted from a pattern of an input character and reference vectors of previously stored target characters, wherein the computer comprises:
• a dictionary storage device in which all characters are classified into plurality of classes based on occurrence frequencies of the target characters and the reference vectors are stored together with classification information;
• distance calculation means for calculating distances between the feature vector and the reference vectors of plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes;
• and
• selection means for selecting a character candidate based on distance calculation as to the reference vectors of the characters belonging to the class having the highest character occurrence frequency when distances calculated by the distance calculation means as to one or plurality of characters belonging to the class having the highest character occurrence frequency are smaller than a predetermined threshold, and
• a list indicating order of average distance of features of a number of reference vectors,
• wherein the computer is programmed such that after features of the feature vector extracted from the input pattern are sorted in accordance with the list in descending order of average distance, first, with respect to features up to an n-th feature and with respect to one of the reference vectors, a distance as to the corresponding features is calculated, (1) the distance is compared with a character candidate selection threshold, (2) if the distance is not larger than the threshold, a distance including next n features is calculated, procedures of (1) and (2) are repeated, and when the distance exceeds the character candidate selection threshold, a comparison with a next reference vector is started.

A character recognition method in which all characters are classified into plurality of classes based on occurrence frequencies of target characters, wherein a computer provided with a dictionary storage device for storing the reference vectors for each of the classes is used, and wherein distances between a feature vector extracted from an input pattern and the reference vectors of previously stored target characters are compared to perform character recognition, the method comprising the steps of:

• (a) calculating distances between the feature vector and the reference vectors of plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes;
• (b) selecting, when there is a distance smaller than a class threshold determined for the class, a character of a reference vector corresponding to the distance as a candidate character, wherein a list indicating order of average distance of features of a number of reference vectors is used;
• (c) sorting features of the feature vector extracted from the input pattern in accordance with the list in descending order of average distance, and then, calculating a distance as to corresponding features with respect to one of the reference vectors and with respect to features up to an n-th feature;
• (d) repeating following steps (d-1) and (d-2), and proceeding to a comparison with a next reference vector when the distance exceeds a character candidate selection threshold,
• (d-1) comparing the distance with the character candidate selection threshold, and
• (d-2) calculating a distance including next n features if the distance is not larger than the threshold.

A character recognition method in which all characters are classified into a plurality of classes based on occurrence frequencies of target characters, wherein a computer provided with a dictionary storage device for storing the reference vectors for each of the classes is used, and wherein distances between a feature vector extracted from an input pattern and the reference vectors of previously stored target characters are compared to perform character recognition, the method comprising the steps of:

• (a) calculating distances between the feature vector and the reference vectors of plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes;
• (b) selecting, when there is a distance smaller than a class threshold determined for the class, a character of a reference vector corresponding to the distance as a candidate character, wherein a similar character dictionary for providing characters having similar relations to arbitrary characters is used;
• (c) taking out characters similar to the respective character candidates from the similar character dictionary in response to a selection of the plurality of character candidates;
• (d) calculating distances between the feature vector to be recognized and reference vectors of the similar characters;
• (e) sorting distance values of the plurality of character candidates and distance values of the similar characters;
• and
• (f) outputting a predetermined number of characters in ascending order of distance as recognition character candidates.

A character recognition method in which all characters are classified into a plurality of classes based on occurrence frequencies of target characters, wherein a computer provided with a dictionary storage device for storing the reference vectors for each of the classes is used, and wherein distances between a feature vector extracted from an input pattern and the reference vectors of previously stored target characters are compared to perform character recognition, the method comprising the steps of:

• (a) calculating distances between the feature vector and the reference vectors of plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes;
• (b) selecting, when there is a distance smaller than a class threshold determined for the class, a character of a reference vector corresponding to the distance as a candidate character;
• (c) calculating, when there is no distance smaller than the class threshold, distances between the feature vector and the reference vectors of plurality of characters belonging to a class having a second highest character occurrence frequency;
• (d) selecting, when there is a distance smaller than a second class threshold determined for the class, a character of a reference vector corresponding to the distance as a candidate character, wherein a similar character dictionary for providing characters having similar relations to arbitrary characters is used;
• (e) taking out characters similar to the respective character candidates from the similar character dictionary in response to a selection of the plurality of character candidates, and calculating distances between the feature vector to be recognized and reference vectors of the similar characters;
• and
• (f) sorting distance values of the plurality of character candidates and distance values of the similar characters, and outputting a predetermined number of characters in ascending order of distance as recognition character candidates.

A computer readable recording medium storing a dictionary in which all characters are classified into plurality of classes based on occurrence frequencies of target characters, and reference vectors are stored for each of the classes, and a computer program executing the steps of:

• (a) calculating distance values between a feature vector extracted from an input pattern and reference vectors of plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes;
• and
• (b) selecting, when there is a distance value smaller than a class threshold determined for the class having the highest character occurrence frequency, a character of a reference vector having the distance as a character candidate, wherein a list indicating order of average distance of features of a number of reference vectors is further stored on said medium,
• (c) sorting features of the feature vector extracted from the input pattern in accordance with the list in descending order of average distance, and then, calculating a distance as to corresponding features with respect to one of the reference vectors and with respect to features up to an n-th feature;
• and
• (d) repeating following steps (d-1) and (d-2), and proceeding to a comparison with a next reference vector when the distance exceeds a character candidate selection threshold,
• (d-1) comparing the distance with the character candidate selection threshold, and
• (d-2) calculating a distance including next n features if the distance is not larger than the threshold.

A computer readable medium that stores a program that causes a computer to perform character recognition, said program comprising instructions that cause said computer to:

• compare distances between a feature vector extracted from an input pattern and reference vectors of previously stored target characters, wherein all characters are classified into plurality of classes based on occurrence frequencies of the target characters,
• access a stored dictionary of the reference vectors for each of the classes,
• to calculate distances between the feature vector and the reference vectors of a plurality of characters belonging to a class having a highest character occurrence frequency among the plurality of classes, and to select, when there is a distance smaller than a class threshold determined for the class, a character of a reference vector corresponding to the distance as a character candidate, and
• to use a similar character dictionary for providing characters having similar relations to arbitrary characters, in taking out characters similar to the respective character candidates from the similar character dictionary in response to a selection of the plurality of character candidates,
• to calculate distances between the feature vector to be recognized and reference vectors of the similar characters,
• to sort distance values of the plurality of character candidates and distance values of the similar characters, and
• to select a predetermined number of characters in ascending order of distance as recognition character candidates.