JPH06103403A - Optical character reader - Google Patents

Abstract

(57) [Abstract] [Purpose] To provide an optical character reading device in which correction work by an operator is reduced and the recognition rate of characters is improved. [Structure] The recognition unit 6 outputs a character candidate string using the recognition dictionary unit 5. The knowledge processing unit 8 collates the combination of character candidates with the word dictionary and outputs the word candidate sequence in the descending order of the score.
When the scores of the first rank and the second rank are the same in the word candidate sequence, the control unit 9 regards the character image which is different between the first rank and the second rank and the common character between the first rank and the second rank The recognition unit 6 is controlled so as to collate with the character patterns stored separately in the recognition dictionary unit 5 and output the character candidates again. In this way, even if the first and second scores in the word candidate string are the same, they are not rejected but are stored separately in the recognition dictionary unit 5 as a character image with a low accuracy rate. Since the character candidates are output again by collating with the character patterns in a narrow range, it is possible to output the character candidates with a high accuracy rate.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character reader.

[0002]

2. Description of the Related Art Conventionally, in an optical character reader, a method called knowledge processing is used when a word to be written on a form is limited in its range of use such as an address and a family name. This is adopted to improve the recognition rate of characters.

This knowledge processing is a method of collating a combination of character candidates obtained by character recognition processing with a word dictionary and outputting word candidates in descending order of score.

However, for example, when the character "Forest" is not well recognized for the surname "Forest" due to the influence of garbage, etc., the beginnings of "Forest" are all like "Kimura", "Forest". The score of the word candidates becomes the same score, which causes the inconvenience that the final word cannot be determined.

Therefore, in such a case, it is conceivable to output the frequently used word candidate "Kimura", but since it may be erroneously read, it is rejected without being output, so that the operator can see that part. Was being corrected.

[0006]

However, although the misreading rate is reduced by rejecting, there is a problem in that the operator may need to input correct characters, which may cause an excessive burden on the operator in some cases.

Therefore, the present invention has been made in view of the above circumstances, and reduces correction work by an operator,
It is an object of the present invention to provide an optical character reading device with an improved character recognition rate.

[0008]

In order to achieve the above object, the optical character reader of the present invention is a recognition dictionary unit in which character patterns forming a word are stored separately for each character common to the words. A scanner unit that detects the image of the form, a character cutout unit that cuts out a character image from the form image, a character image cut out from the form image, and all the character patterns stored in the recognition dictionary unit. A recognition unit that outputs a character candidate string, a knowledge processing unit that collates the combination of character candidates output by this recognition unit with a word dictionary, and outputs a word candidate string in descending order of the score, and a knowledge processing unit that outputs this word processing unit. If the scores of the first and second places in the word candidate sequence are the same, the division in the recognition dictionary unit in which the character pattern is stored in association with the character common between the first and second places is designated. , First , A control unit for controlling the recognition unit so as to collate the character image having different characters between the second place and the character pattern in the designated section and re-output the character candidate. To do.

[0009]

The operation of the optical character reader having the above construction will be described.

When the scanner unit detects the image of the form, the character cutout unit cuts out the character image from the form image. The recognition unit collates the cut out character image with all the character patterns stored in the recognition dictionary unit, and outputs a character candidate string. The knowledge processing unit compares the output character candidate combination with the word dictionary to obtain a score, and then outputs a word candidate string in the descending order of the score.

Here, the control unit determines whether or not the scores of the first rank and the second rank of the output word candidate strings are the same. Then, in the case of the same point, the control unit designates the division in the recognition dictionary unit in which the character pattern is stored in association with the character common between the first and second places. Under the control of the control unit, the recognition unit collates a character image that is a different character between the first and second places with a character pattern in the section designated by the control unit, and re-outputs the character candidate.

As described above, even if the first and second points in the word candidate string have the same score, they are not rejected but are divided into a character image having a low correct answer rate and a recognition dictionary section. Since the stored character patterns are collated and the character candidates are output again, it is possible to output the character candidates with a high accuracy rate.

[0013]

Embodiments of the present invention will be described below in detail with reference to the drawings.

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

The apparatus 1 of this embodiment shown in FIG. 1 includes a scanner section 2 for detecting an image of a form P, a sheet buffer 3 for storing a form image detected by the scanner section 2,
1 from the form image stored in this sheet buffer 3
A character slicing unit 4 for slicing out a character image for each character, and a recognizing unit 6 for performing a character recognizing process by collating a character image cut out by the character slicing unit 4 with a dictionary stored in a recognizing dictionary unit 5, and a recognizing unit. 6 recognition result and word dictionary unit 7
A knowledge processing unit 8 for performing knowledge processing by collating with a dictionary stored in and a control unit 9 for controlling each unit of the device 1 are roughly configured.

Next, the details of each of the above parts will be described.

The scanner unit 2 includes a light source for irradiating the form P with light and a photoelectric conversion element for receiving the reflected light from the form P and converting it into an electric signal, and optically scans the entire form P. The form image is detected.

In the recognition dictionary section 5, character patterns forming a word are stored in each partial recognition dictionary 5a which is divided for each character common to the words. For example, in one partial recognition dictionary 5a, the character patterns of the second character (second character and third character) that make up the family name and the first character is "tree" are common to the words " It is stored in relation to "tree". Further, the other partial recognition dictionary 5a stores all the character patterns forming the first and last names and all the character patterns forming the address. The character patterns stored in the dictionary unit 5 are stored in the form of characteristic information of each character necessary for the character recognition processing by the recognition unit 6.

The recognition unit 6 has a buffer memory, stores the character image cut out by the character cutting unit 4 in the buffer memory, and performs recognition processing, that is, pattern matching processing on the stored character image, This is to output an example of character candidates. In the pattern matching processing performed here, the similarity value of the character image and all or part of the character pattern stored in the recognition dictionary unit 5 is calculated by a composite similarity method, and the first similarity value is calculated in descending order. The nth place from the 1st place (eg 15th place)
Up to a plurality of character candidates (character codes) are determined. Then, such a pattern matching process is performed at the time of the first recognition process using all the partial recognition dictionaries 5a and the re-recognition process using the specific partial recognition dictionaries 5a.

The knowledge processing unit 8 gives ranking points according to the ranks of the character candidates output from the recognition unit 6, and a candidate composed of a plurality of character candidate strings obtained by combining the plurality of character candidates in a matrix format. A character matrix is generated, these character candidate strings are collated with the dictionary stored in the word dictionary unit 7, the score as a word candidate is obtained as the sum of the ranking points of each character candidate, and the word candidate string is arranged in descending order of the score. Is determined and the knowledge processing result is output to the control unit 9. The ranking points are, for example, 10th for the first ranking.
0 points, 2nd place are given 90 points, 3rd place are given 80 points, and 0 points are given when there is no character candidate.

Based on the knowledge processing result output from the knowledge processing unit 8, the control unit 9 determines whether or not the scores of the first and second positions in the word candidate sequence are the same. If they are not the same, the knowledge processing unit 8 outputs the first-ranked word candidate to the output unit (not shown).
A partial recognition dictionary 5a in the recognition dictionary unit 5 in which a character pattern is stored in association with a character common between ranks is designated, and a character image in which different characters are placed between the first and second ranks and the designated character image are designated. The recognition unit 6 is controlled so as to collate with the character pattern in the partial recognition dictionary 5a and re-output the character candidate.

Next, the operation of this embodiment will be described with reference to the flowchart of FIG.

It is assumed that the form P is filled with "Kirin" which is the family name.

When the scanner unit 2 optically scans the entire form P under the control of the control unit 9 and detects a form image, the form image “Kibayashi” is output to the sheet buffer 3. The sheet buffer 3 stores the form image “Kibayashi” output from the scanner unit 2 (S1).

Subsequently, the character cutout unit 4 cuts out the character images "wood" and "bay" for each character from the form image "wood" stored in the sheet buffer 3 and outputs them to the recognition unit 6 (S2).

Next, the recognition unit 6 stores the character images "tree" and "bay" output from the character cutout unit 4 in a buffer memory, and recognizes the character images "tree" and "bay" in the recognition dictionary unit 5 as well. All partial recognition dictionaries 5 stored in
The recognition processing is performed using a, the similarity value is obtained, and the character candidate strings from the first rank to the nth rank (fifteenth) are output to the knowledge processing unit 8 in descending order of similarity value (S3). in this case,
As shown in FIG. 2, as the character candidate string for the character image “tree”, the first place “tree”, the second place “rice”, the third place “not”, ... Are output, and the character image “bay” is produced. As the character candidate string, the first rank “o”, the second rank “age”, ... Are output, and the “village” and “bay” are not output.

Then, the knowledge processing unit 8 gives ranking points according to the ranking of the character candidates output from the recognition unit 6. The ranking points in this case are, as shown in FIG. 2, 100 points for the first character candidate “tree” and “O”, and the second character candidate “rice”,
The "age" is 90 points, and the third character candidate "not" is 80 points. Subsequently, the knowledge processing unit 8 generates a candidate character matrix composed of a plurality of character candidate strings in which the plurality of character candidates are combined in a matrix format, and these character candidate strings and a dictionary stored in the word dictionary unit 7 are generated. Match
The score as a word candidate is obtained as the sum of the ranking points of each character candidate, the word candidate string is determined in the descending order of the score, and the knowledge processing result is output to the control unit 9 (S4). In the word candidate string in this case, as shown in FIG. 2, the first rank is “Kimura” and the second rank is
It is assumed that “Kibayashi” is output as the rank, and “Kinoshita” is output as the 15th rank. The score of each word candidate is 0 if the second character candidate “mura”, “bayashi”, ..., “Down” has no candidate and the character candidate “tree” has 100 points. Add each to get 100 points.

Next, the control unit 9 determines whether or not the scores of the first rank and the second rank of the word candidate strings are the same based on the knowledge processing result of the knowledge processing unit 8 (S5). Otherwise, the first word candidate is output as a final word candidate from the knowledge processing unit 8 to an output unit (not shown).

In the example shown in FIG. 2, since the scores of the first place and the second place are both 100 points, the control section 9 controls the first and second places.
The partial recognition dictionary 5a in the recognition dictionary unit 5 in which the character pattern "Hayashi" is stored as the second character of the surname in association with the character common to the ranks is designated, and the recognition unit 6 is made to perform the re-recognition process. (S6). Under the control of the control unit 9, the recognition unit 6 collates the character image “Hayashi” which has different characters between the first and second places with the character pattern in the partial recognition dictionary 5a designated by the control unit 9. Then, the similarity value is obtained in the same manner as described above, and the character candidate strings are arranged in the descending order of the similarity value in the knowledge processing unit 8.
Output again. In this case, as shown in FIG. 2, it is assumed that the first rank “bay”, the second rank “mura”, ... Are output as the character candidate strings for the character image “bay”.

Under the control of the control unit 9, the knowledge processing unit 8
A plurality of character candidates of the character candidate string for the second character image “Hayashi” re-output by the recognition unit 6 and the character candidate string of the first character already sent from the recognition unit 6 are combined in a matrix format to form candidates. A character matrix is generated, these character candidate strings are collated with the dictionary stored in the word dictionary unit 7, the score as a word candidate is obtained as the sum of the ranking points of each character candidate, and the word candidate string is arranged in descending order of the score. Is determined and output to an output unit (not shown) (S7). In this case, the first-ranked “Kibayashi”, the second-ranked “Kimura”, ... Are output to the output unit, and the correct word “Kibayashi” is obtained.

According to the above-described embodiment, even if the first and second ranks of the word candidate strings output by the knowledge processing unit 8 have the same score, the correct answer is not rejected. Since a character image having a low rate and a character pattern in the partial recognition dictionary 5a in a narrow range which is divided and stored in the recognition dictionary section 5 are collated and a character candidate is output again, a character having a high correct rate is obtained. It is possible to output candidates, reduce the correction work by the operator, and provide the optical character reading device 1 in which the recognition rate of characters is improved.

The present invention is not limited to the above embodiment,
Various modifications can be made without changing the gist of the invention.
For example, the recognition process performed first by the recognition unit 6 may use not the partial recognition dictionaries 5a for all, but the partial recognition dictionaries 5a for the first and last names designated as the format control from the control unit 9.

[0033]

According to the present invention described in detail above, even if the first and second points in a word candidate string have the same score, they are not rejected and a character image with a low correct answer rate is obtained. Since it is designed to output character candidates again by collating with a narrow range of character patterns stored separately in the recognition dictionary section, it is possible to output character candidates with a high accuracy rate.
It is possible to provide an optical character reading device that reduces the correction work by the operator and improves the character recognition rate.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing the operation of this embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Optical character reader 5 Recognition dictionary part 5a Partial recognition dictionary 6 Recognition part 7 Word dictionary part 8 Knowledge processing part 9 Control part

Claims (1)

[Claims] 1. A recognition dictionary unit in which character patterns forming a word are divided and stored for each character common to words, a character image cut out from a form image, and all of the recognition dictionary units stored therein. A recognition unit that collates a character pattern and outputs a character candidate string, and a knowledge processing unit that collates the combination of character candidates output by this recognition unit with a word dictionary and outputs a word candidate string in descending order of score, When the first and second ranks of the word candidate strings output by the knowledge processing unit have the same score, the recognition in which the character pattern is stored in association with the character common between the first and second ranks. The recognition unit is controlled to specify a section in the dictionary section, collate the character image having different characters between the first and second places with the character pattern in the specified section, and re-output a character candidate. To have a control unit that An optical character reader characterized by.