JPH06176206A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve accuracy for correct answers and to accelerate processing speed in the case of OCR knowledge processing. CONSTITUTION:A character recognition part 3 recognizes character images such as 'last names' and 'first names' recorded in plural fields and provides plural recognized result candidates. A knowledge processing part 7 narrows down the recognized result candidates of the character recognition part 3 corresponding to the prescribed field such as the 'first name' based on image identification information in the other specified field like a sex identification mark 2A, for example. Namely, the knowledge processing part 7 collates the sex shown by the sex identification mark 2A with sex flags respectively added to 'first name' dictionary data 6B in a knowledge dictionary memory 5 loading the last name dictionary data 6B of a knowledge dictionary file 6 and performs similarity collation and judge processing only to the candidate with the coincident sex.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device in a computer system.

[0002]

2. Description of the Related Art A general Kanji OCR device outputs a plurality of candidate characters from a recognition unit for the purpose of compensating for the lack of basic character recognition ability for each character. That is, I think it's probably "A", but as I said, it might be "B", "C", ... 1
The result of one read character is not uniquely output as one character, but a plurality of results are output in descending order of possibility. And then
The optimum character is selected from these candidate characters. As described above, a general OCR device uses a two-step method for improving the apparent recognition rate.

A typical process used as the latter process is the knowledge process. This is a method of selecting a character from candidate characters according to the degree of similarity of words in the dictionary by comparing with a dictionary of "words" in which a plurality of characters are combined.

For example, in OCR V-3000 manufactured by Toshiba Corp. as shown in FIG. 5A, when the OCR recognizes a character, the character recognition unit 100 has a maximum of 15 (ANK character ... Candidate characters (1st to nth) of maximum 3, kanji ... maximum 15) are selected from the recognition dictionary 102.
The knowledge processing unit 104 weights candidate characters given in character units in the order of candidates. The weight is set such that the first-ranked candidate character is "10", and the second-ranked characters are "9", "8", ..., "2", "1" (minimum "1"). Next, the word dictionary registered in the knowledge base dictionary 106 (actually, the knowledge dictionary memory 108 when the power is turned on).
(Word dictionary loaded in), and match (total weight) in word units, and select words in descending order of similarity. At this time, words with different word lengths (number of characters) are excluded. From the selected word candidates, the word that exceeds the threshold value registered in the knowledge base dictionary 106 in advance and has the highest similarity is output as the recognition result. At this time, if there is no word exceeding the threshold value, the character recognition unit 100 outputs the first candidate.

For example, in the case shown in FIG.
The candidate characters of the plain reading result of “Oita prefecture” on the form 110 are displayed from the first to the fifteenth place, and the weight is set for each (10 points to 1 point). Oita Prefecture word length (word character count)
Is "3", so only words with a word length of 3 among the words registered in the knowledge dictionary are searched. The sum of weight values is calculated for each word, and the value obtained as the result is used as the score for that word. As a result of the calculation, "Oita prefecture" having the highest score becomes the word having the highest degree of similarity, but "Oita prefecture" is finally output only when the threshold value is exceeded.

[0006] Here, in order to make the threshold value easy to understand,
Think in "%". In the above case, since the threshold value is "06", if the weight calculation result is 60% or more of the maximum weight, the answer is given. First, each character is given 10 points (highest weight value). In the above example, there are 3 characters, so 3
It becomes 0 points. If the threshold is "06 (60%)", 1
Eight points (10 points x 3 characters x 0.6) are candidates for an answer.
If there are multiple words with 18 points or more, the one with the highest score is the answer. In addition, when there are a plurality of items with the highest scores, the word with the highest registration rank in the knowledge dictionary is given priority.

[0007]

Although the apparent recognition rate of the Kanji OCR device has been dramatically improved by adopting the two-step method as described above, it cannot be said to be perfect yet and the reading rate is further improved. Improvement of (correct answer accuracy) and improvement of processing speed are issues.

An object of the present invention is to improve the accuracy of correct answer and the processing speed in the OCR knowledge processing.

[0009]

The means of the present invention are as follows. The recognition means recognizes the character images recorded in the plurality of fields. The narrowing-down unit narrows down the recognition result candidates of the recognition unit for the predetermined field based on the image identification information in another specific field.

[0010]

The operation of the means of the present invention is as follows. First,
The recognition means recognizes character images recorded in a plurality of fields such as "surname" and "first name", and obtains recognition result candidates respectively. Then, the narrowing-down means narrows down the recognition result candidates of the recognizing means for a predetermined field such as "first name" based on image identification information in another specific field such as another identification mark of male or female. That is, for example, paying attention to the distinction between male and female of "first name", read the identification mark of the male and female etc. (○ mark etc.) together with the reading of the surname and first name, and use this male and female identification data. After selecting the names in the dictionary by gender, the correct answer rate is increased and the processing speed is improved by selecting the one with the highest similarity from the narrowed down list.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT An embodiment of the present invention will be described below with reference to FIGS.

FIG. 1 is a view showing the structure thereof. Reference numeral 1 in the figure is an image scanner unit for reading handwritten characters on the input form 2, and 3 is a character read by the image scanner unit 1. A character recognition unit that obtains similar candidate characters from the recognition dictionary configured in the recognition dictionary memory 4. Reference numeral 5 is a knowledge dictionary memory in which the contents of the knowledge base dictionary provided in the form of the knowledge dictionary file 6 are loaded at system startup. Reference numeral 7 is a knowledge processing unit for selecting one of the candidate characters output from the character recognition unit 3 by collating the knowledge dictionary memory 5, and 8 is not shown as the read result data for the selected candidate character. It is an output interface unit that outputs to a computer. Note that 3A and 7A in the figure are work memories.

The knowledge dictionary file 6 is composed of, for example, surname dictionary data 6A, first name dictionary data 6B, product name dictionary data 6C, and the like. Is attached to a male flag and a female flag indicating to be used by a female (both flags are collectively referred to as a gender flag).

Further, the input form 2 is provided with a "sex" field in which a sex identification mark (○ mark) 2A is entered in addition to the entry fields such as "surname", "first name" and "product name". Since this "sex" field needs to be processed before the "first name" field, it is arranged at a position where it can be read by the image scanner unit 1 before the "first name" field.

Since there is also a "name" for both men and women, for such a "name" word, both the male and female flags in the name dictionary data 6B are set and the "sex" read.
Whether the field is male or female, it will be the target of verification.

Next, the operation in such a configuration will be described.

First, when the system is started up, FIG.
As shown in, the contents of the knowledge base dictionary provided in the form of the knowledge dictionary file 6 are set as system initial settings.
Prior to processing, the knowledge dictionary memory 5 is loaded (step S11).

When the form 2 is read, the processes shown in FIGS. 3 and 4 are performed. That is, first, the image scanner unit 1 reads the form 2 and outputs the image data to the character recognition unit 3 (step S21). The character recognition unit 3 cuts out a character pattern from the output data of the image scanner 1 character by character and temporarily stores it in the work memory 3A (step S22). Next, the character patterns temporarily stored in the work memory 3A are collated character by character with the basic character patterns of the recognition dictionary 4 for similarity (step S2).
3). Then, a plurality of character codes as candidate characters are output as recognition results in descending order of similarity (step S2).
4). The matching of the degree of similarity and the output of the recognition result are performed for all the character data temporarily stored in the work memory 3A (step S25).

The knowledge processing unit 7 sequentially stores the characters output from the character recognition unit 3 in the work memory 7A. Then, the characters are grouped in field units (word units), and 1
When the data for the fields is collected, the word dictionary in the knowledge dictionary memory 5 is compared. That is, the work memory 7A
Then, the characters for one field (word) are extracted (step S26), and it is determined whether the field is the "sex" field (step S27). In the example of FIG. 1, since the "surname" field is read first, the result is "No", and then the process proceeds to the determination of whether the field is a knowledge processing target field (step 28). Here, since the "surname" field is the knowledge processing target field, it becomes "Yes", and the process proceeds to the determination of whether or not it is the next "first name" field (step S29). Next time,
Since it is not the "first name" field, the result is "No", and the normal dictionary matching process similar to the conventional one, in this case, the matching process with the surname dictionary data 6A is performed (step S30). As a result of this matching, the word with the highest degree of similarity is output as the recognition result (step S31). If the degree of similarity does not reach a certain level, the first rank character of the recognition result is output as a word not in the dictionary. If there are still unprocessed characters in the work memory 7A, the above step S2 is performed.
Repeat from 6.

That is, in step S26,
When the characters in the "sex" field are extracted after the "surname" field, the determination in step S27 is "Yes".
Becomes As described above, in the case of the gender identification mark 2A on the "sex" field, the gender flag indicating whether the "name" is male or female is added to the knowledge processing unit work memory 7A.
(Step S33).

In the collation process after reading the "name" field, "Yes" is obtained in step S29, and the "name" dictionary data 6B in the knowledge dictionary memory 5 is previously stored in the knowledge dictionary memory 5 based on the previously read gender identification flag. Matching is carried out with the respective gender flags, and only the matching gender flags are subjected to similarity matching and determination processing (step S3).
4).

With this processing, for example, in the example of FIG. 1, when the "name" field is read, a plurality of characters are read for reading one character from the character recognition unit 3 as shown in FIG. Although the candidate character of is output, the knowledge processing unit 7
In the above, since only the man flag is set at the time of matching with the knowledge dictionary memory 5, it is possible to quickly select the most accurate word "hajime" as shown in FIG. 2C. It will be possible. Note that FIG.
(D) indicates output data for the form 2 in the example of FIG.

As described above, the character recognition unit 3 recognizes character images recorded in a plurality of fields such as "surname" and "first name", obtains recognition result candidates, and the knowledge processing unit 7 determines, for example, "first name". The recognition result candidates of the character recognition unit 3 for a predetermined field such as "" are narrowed down based on image identification information in another specific field such as another identification mark of male / female. It is possible to improve the accuracy of the correct answer and the processing speed.

The present embodiment can be applied to any "name" processing of reading kanji, reading kana, and reading English.

In the above embodiment, the recognition result candidates of the character recognition unit 3 for a predetermined field such as "name" are
Although the narrowing down is performed based on the image identification information in other specific fields such as different identification marks for male and female, this predetermined field and other specific fields are not limited to the above-mentioned embodiment.

[0026]

According to the present invention, a flag indicating the image recognition result of a specific field is provided in the knowledge dictionary, and character recognition result candidates for a predetermined field are narrowed down based on this flag. It is possible to efficiently narrow down, and it is possible to improve the correct answer rate and processing speed of knowledge processing.

[Brief description of drawings]

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

2A is an operation flowchart at system startup, FIG. 2B is an output example of a character recognition unit, FIG. 2C is an output example of a knowledge processing unit, and FIG. 2D is an output result from an output interface unit. It is a figure which respectively shows.

FIG. 3 is a first half portion of a series of operation flowcharts at the time of reading a form.

FIG. 4 is the latter half of a series of operation flowcharts for reading a form.

5A is a block configuration diagram of a conventional character recognition device, and FIG. 5B is a diagram for explaining an operation of the conventional character recognition device.

[Explanation of symbols]

1 ... Image scanner part, 2 ... Input form, 2A ... Gender identification mark (○ mark), 3 ... Character recognition part, 4 ... Recognition dictionary memory, 5 ... Knowledge dictionary memory, 6 ... Knowledge dictionary file, 6A
... family name dictionary data, 6B ... name dictionary data, 6C ... product name dictionary data, 7 ... knowledge processing section, 8 ... output interface section.

Claims (1)

[Claims] 1. A recognition means for recognizing a character image recorded in a plurality of fields, and a narrowing means for narrowing down recognition result candidates of the recognition means for a predetermined field based on image identification information in another specific field. A character recognition device characterized by: