JPH076212A - Intelligence processing unit for optical character reader - Google Patents

Abstract

PURPOSE:To provide the intelligence processing unit for the optical character reader in which processing time is short and characters are surely recognized. CONSTITUTION:The processing unit is provided with a font-depending word dictionary 8 for each font, a font retrieval section 6 retrieving a character string comprising a same font from an object character group for each font from a recognition section 4 and calculating the priority of the character string from the object order of the object characters being components of the character string, and a collation section 7 collating the font-depending word dictionary 8 for each font of the same kind as the font of the character string retrieved depending on the priority and providing an output of a read result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge processing device for an optical character reader, and more particularly to a plurality of character types (Kanji, Hiragana,
The present invention relates to a knowledge processing device of an optical character reading device that performs word matching using a dictionary from a candidate character group obtained by reading a general sentence in which characters such as katakana, numbers, and English letters are mixed.

[0002]

2. Description of the Related Art There is an optical character reader for reading a character written on a medium such as paper and recognizing the character, but it is possible to perform complete recognition only by pattern-processing the image data of the written character. Is difficult, and it is necessary to improve the recognition rate by performing some knowledge processing after reading the characters.

In the knowledge processing of the conventional optical character reader,
We prepare various dictionaries such as address dictionary, first and last name dictionary, company name dictionary, job title dictionary, apartment / condominium name dictionary, occupation dictionary, and disease name dictionary. The information is set and registered in advance in the device, and after reading the characters, the corresponding dictionary is collated for each entry field to determine the characters. In the example shown in FIG. 8, each of the company name, job title, and name fields is provided on the entry form 51, and the knowledge of the company name dictionary, job name dictionary, and surname family dictionary for the character strings entered in these fields. It is designed to be matched with each dictionary.

When the character string to be read is a general sentence, all the above-mentioned dictionaries including the general word dictionary are used. In addition, a plurality of candidate characters is obtained for each character of the character string to be read and set as a candidate character group. Then, as shown in FIG. 9, the first candidate character for the first character (first character) of the character string in the candidate character group is read, and the word having the first candidate character as the first character is read from the dictionary. A search is performed and a comparison is made as to whether the second and subsequent characters of the word are present after the second and subsequent characters of the candidate character group.
If the word can be collated by comparison, the word is output as a read result, and if the word cannot be collated, the dictionary side word to be collated is updated, and the collation processing is repeated.

As a result of matching with a series of words whose first character is the first candidate character of the first character of the candidate character group as a result of the matching process described above, if no matching word is found, the candidate character is selected. The same matching process is executed for the word having the first candidate character of the first character of the group as the first character.

For all the first candidate characters of the candidate character group, if no word having the first character as the first character is found by word matching, the word having the second character of the candidate character group as the first character is found. The above-mentioned processing is executed in order to collate. Then, these processes are executed for all the characters of the character string to be read, and the result is output.

Further, Japanese Patent Publication No. 1-19195 discloses that a candidate character is weighted according to a recognition rank, and when matching with a word, the word having the highest weight value is selected and output in consideration of the weight. A structured knowledge processing method is disclosed.

[0008]

Generally, in Japanese sentences, characters of a plurality of character types (Kanji, Hiragana, Katakana, numbers, English letters, etc.) are mixed. However, in the conventional knowledge processing method described above, Even if the first character and the second character of the portion to be collated are of different character types, the collation processing is performed, and there is a problem that unnecessary processing time is required. In addition, since word matching is performed without being aware of the fact that the character types are different, there is a case where the character should not be originally extracted as a word, for example, when characters of multiple character types are listed as candidate characters. Nevertheless, there is a problem that it may be recognized as a word.

It is an object of the present invention to provide a knowledge processing device for an optical character reading device which requires a short processing time and can reliably perform character recognition.

[0010]

A knowledge processing device of an optical character reader according to the present invention comprises a scanning unit for irradiating a medium on which characters are written with light and converting reflected light from the medium into an electric signal for storage. And a pre-processing unit that cuts out the stored electric signal in units of one character to adjust the size, and cut out one
A feature extraction unit that extracts features for each character, a recognition dictionary,
In a knowledge processing device used in an optical character reading device having a recognition unit that compares extracted features with the contents of the recognition dictionary and outputs a plurality of candidate characters for each extracted character as a recognition result. , A character type word dictionary provided for each character type, and a character string having the same character type for each character type are searched from a candidate character group output from the recognition unit, and candidate character candidates that form the character string are searched. And a collation unit that collates a character type word dictionary of the same character type as the character type of the character string according to the priority and outputs a reading result. Have.

[0011]

[Function] A character string consisting of the same character type is searched from the candidate character group for each character type, the priority order for this character string is calculated from the candidate character candidate ranks, and the character type word dictionary provided for each character type is calculated. Since the reading result is obtained by collating, the range of words on the dictionary side to be subjected to a series of collations can be narrowed down, and the time required for the collation processing can be shortened. Further, since word recognition is performed according to the division between different character types, erroneous recognition is reduced and the recognition rate is improved.

[0012]

Embodiments of the present invention will now be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an optical character reading device according to an embodiment of the present invention. This optical character reading device has a search function for a character string of the same character type.

A scanning unit 1 is provided which irradiates a medium such as paper on which characters are written with light and converts reflected light from the medium into an electric signal and stores the electric signal. The scanning unit 1 cuts the stored electric signal in units of one character. A preprocessing unit 2 that adjusts the size of the operation (performs normalization) is connected to the scanning unit 1. A feature extraction unit 3 that extracts the feature of each character normalized by the preprocessing unit 2 and a recognition dictionary 5 that stores the relationship between the character and the feature.
And are provided. A recognition unit 4 that receives the output of the feature extraction unit 3 and searches the recognition dictionary is provided, and the recognition unit 4 compares the contents of the extracted feature and the recognition dictionary 5,
A plurality of candidate characters is selected for each cut out character and is output as a recognition result.

Furthermore, this optical character reader is designed to
A plurality of character type word dictionaries 8 provided for each character type such as hiragana, katakana, letters, numbers, and symbols, and a character type search for a candidate character group output from the recognition unit 4 are performed to obtain a character string. For the character type search unit 6 to be output and the character string output from the character type search unit 6, a character type word dictionary of the same character type as the character type of the character string is collated according to the priority of the character string. The verification unit 7 outputs the reading result. The character type search unit 6 searches the candidate character group from the recognition unit 4 for a character string having the same character type for each character type, and prioritizes the character string from the candidate ranks of the candidate characters forming the character string. It is configured to calculate ranks.

Next, the operation of this embodiment will be described.

The text entered on a paper or the like is read by the scanning unit 1, cut out by the pre-processing unit 2 for each character and adjusted in size, and the feature extraction unit 3 extracts the features for each character. , To the recognition unit 4. The recognition unit 4 collates with the recognition dictionary 5, and a plurality of candidate characters are selected and output for each cut out character. Each candidate character is assigned a candidate rank. The processing up to this point is similar to that of the conventional optical character reading device.

Below, the 18-character "English character END"
Is read in katakana ”
The recognition unit 4 will be described assuming that each character outputs nine candidate characters. As a result, as shown in FIG. 2, a candidate character group consisting of 18 × 9 = 162 candidate characters is output from the recognition unit 4. The read character No. indicates the number of the character from the beginning of the sentence in the read character string.

Subsequently, the above-mentioned candidate character group is input to the character type search unit 6 and the same character type search is performed. In the same-character-type search, first, a character type is distinguished for each character of the candidate character for each read character No., and the sum of the number of characters and the weight is obtained for each character type. The weight is 9 when the recognition rank is the first rank, 8 when the recognition rank is the second rank, and 1 in the same manner when the recognition rank is the ninth rank.

In the example described here, the read character No. 1
As for each of the candidate characters, “Ei Aoi Fu Mosho Ganryokuka” is all kanji, the number of kanji is 9 and the weight is 45. Similarly, for read character No. 3, the number of hiragana is 3, the weight is 17, the number of kanji is 5, and the weight is 2.
0, the number of symbols is 1, and the weight is 8.

After the above-described processing is performed for each read character, it is checked for each character type whether or not the characters of that character type are continuously present in the candidate characters of each read character in the order of the read character number. . If they exist consecutively, this is regarded as the same character type string, and the sum of the number of characters and the weight of the same character type in the candidate characters is obtained for each same character type string. In the example shown in the figure, there is an alphabetic character in each of the read character No. 4, read character No. 5, and read character No. 6 candidate characters, and read character No. 3 and read character No.
No English character exists in the candidate character No. 7, so the read character N
Alphabet string N with o.4, read character No. 5 and read character No. 6
o.1. Then, the total number of letters in these read characters and the sum of weights for the letters are obtained. In the example shown here, the total number of alphabetic characters is 4 and the sum of weights is 31, as shown in the same character type string search result column in FIG.

FIG. 2 shows the result of the above processing for all the alphabetic characters, hiragana, katakana, kanji, numbers and symbols.
Is shown in. For numbers and symbols, 2
Since the above read character numbers do not appear consecutively, they do not form the same character type string. Therefore, "()" is added to the column of the same character type string search result.

Next, the priority order for each of the same character type strings is obtained from the same character type string search results. Here, the priority order is obtained by comparing the quotient obtained by dividing the weight by the quantity (the number of characters) for each of the same character type strings. For the English character string No. 1, the sum of weights divided by the total number of characters is approximately 7.75. Similarly, for other strings of the same character type, the weight per character is calculated,
Priorities are assigned in descending order of weight per character. The result is shown in the column of the same character type string priority.

When the priority order is obtained for each of the same character type strings, the collating unit 7 selects the read character No. that constitutes the same character type string for the same character type string having the first priority. A character is searched by collating the character with the character type word dictionary 8 of the corresponding character type. In the example shown here, the alphabet string No. 1 is the first rank, so each of the read character No. 4, read character No. 5 and read character No. 6 candidates corresponding to this alphabet string No. 1 The letters are checked against the English word dictionary, and as a result, "E
The English word "ND" matches.

If there is no matching word in the same character type string having the first priority, the dictionary has the same character type string having the second priority (in this case, hiragana string No. 4). Is compared with the corresponding character-type word dictionary. The matching is repeated for the same character type sequence having a lower priority in sequence until a matching word is found.

When a matching word is found, the candidate character group excluding the read character corresponding to the matching word is processed again from the same character type string search as shown in FIG. The word matching is performed in order from the same character type string having a high Hereinafter, the same character type string search and word matching are repeated until there is no same character type string. Figure 4
6 to 9 show each step of this repetition.

At the stage where there is no character string to be collated, single characters remain in a non-collated state. For such a single character string, for example, grammar processing is performed and the final reading result is output. FIG. 7 shows a state when only a single character remains, and as a final reading result, "the English character END is expressed in katakana" is obtained. Shows.

[0027]

As described above, according to the present invention, a character type word dictionary for each character type and a character string having the same character type for each character type are searched from a candidate character group, and a candidate forming the character string is searched. The character type search unit that calculates the priority order of the character string from the candidate order of the character and the character type word dictionary of the same character type as the character type of the searched character string according to the priority order are collated and the read result is output. By providing the matching unit, it is possible to shorten the time required for the matching process until a matching word is found, reduce false recognition, and improve the recognition rate.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an optical character reading device according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a process of character type search (first time) and word matching.

FIG. 3 is a diagram illustrating a process of character type search (second time) and word matching.

FIG. 4 is a diagram illustrating a process of character type search (third time) and word matching.

FIG. 5 is a diagram illustrating a process of character type search (fourth time) and word matching.

FIG. 6 is a diagram illustrating a process of character type search (fifth time) and word matching.

FIG. 7 is a diagram illustrating a process of character type search (sixth time) and word matching.

FIG. 8 is a diagram illustrating a knowledge processing method of a conventional optical character reading device.

FIG. 9 is a diagram illustrating a knowledge processing method of a conventional optical character reading device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Scanning section 2 Pre-processing section 3 Feature extraction section 4 Recognition section 5 Recognition dictionary 6 Character classification search section 7 Collation section 8 Character classification word dictionary

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] November 9, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be amended] Claims

[Correction method] Change

[Correction content]

[Claims]

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0010

[Correction method] Change

[Correction content]

[0010]

A knowledge processing device for an optical character reader according to the present invention comprises a scanning unit for irradiating a medium on which characters are written with light and converting response light from the medium into an electric signal for storage. And a pre-processing unit that cuts out the stored electric signal in units of one character to adjust the size, and cut out one
A feature extraction unit that extracts features for each character, a recognition dictionary,
In a knowledge processing device used in an optical character reading device having a recognition unit that compares extracted features with the contents of the recognition dictionary and outputs a plurality of candidate characters for each extracted character as a recognition result. , A character type word dictionary provided for each character type, and a character string having the same character type for each character type are searched from a candidate character group output from the recognition unit, and candidate character candidates that form the character string are searched. And a collation unit that collates a character type word dictionary of the same character type as the character type of the character string according to the priority and outputs a reading result. Have.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction content]

[0011]

[Function] A character string consisting of the same character type is searched from the candidate character group for each character type, the priority order for this character string is calculated from the candidate character candidate ranks, and the character type word dictionary provided for each character type is calculated. Since the reading result is obtained by collating, the range of words on the dictionary side to be subjected to a series of collations can be narrowed down, and the time required for the collation processing can be shortened. Further, since word recognition is performed according to the division between different character types, erroneous recognition is reduced and the recognition rate is improved. As a method of calculating the priority,
For example, there is a method based on a weighted sum of candidate ranks. It
In addition, if unmatched single characters remain, grammar processing is performed.
You may make it use together.

Claims (1)

[Claims] 1. A scanning unit for irradiating a medium on which characters are written with light and converting reflected light from the medium into an electric signal and storing the electric signal, and a size for cutting out the stored electric signal in units of one character. A preprocessing unit for adjusting the size, a feature extraction unit for extracting features for each extracted character, a recognition dictionary, and the extracted features by comparing the extracted features with the contents of the recognition dictionary.
In a knowledge processing device used in an optical character reading device having a recognition unit for outputting a plurality of candidate characters for each character as a recognition result, a character type word dictionary provided for each character type and output from the recognition unit. A character type search unit that searches a character string having the same character type for each character type from the selected candidate character group, and calculates the priority order of the character string from the candidate order of the candidate characters that form the character string; A knowledge processing device for an optical character reading device, comprising: a collating unit that collates a character type word dictionary of the same character type with a character type of the character string according to a rank and outputs a reading result.