JPH07141471A - Character recognizing method - Google Patents

Abstract

PURPOSE:To provide a character recognizing method which is capable of lowering rejection rate and improving correct recognition rate. CONSTITUTION:In a character recognizing device recognizing an inputted character, a step performing a matching of an inputted character and a candidate corresponding to the inputted character and having a numeric value showing the similarity degree with the inputted character and a step rejecting the candidate for which the matching is performed when the candidate does not have a prescribed numeric value and deciding the candidate not so as to be rejected when it has the prescribed numeric value, are included.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition method, and more specifically, it creates a table based on a similarity distribution or a distance value distribution of erroneous recognition in a character recognition device, and refers to this table. The present invention relates to a character recognition method for performing rejection judgment.

[0002]

2. Description of the Related Art In a conventional character recognition method, a threshold value is set and a difference between a first-rank similarity and a second-rank similarity is calculated.
Alternatively, a character is recognized by performing a rejection determination by comparing the ratio between the first-rank similarity and the second-rank similarity with a predetermined threshold value.

[0003]

[Problems to be Solved by the Invention] The conventional first degree similarity and 2
The simple calculation of comparing the difference between the degree of similarity or the ratio of the degree of similarity of 1st place and the degree of similarity of 2nd place with a predetermined threshold can handle the complicated distribution of misrecognitions to be rejected in detail. However, the rate of rejecting even correctly recognized objects was high, and as a result, the rejection rate was high and the character recognition rate was low. An object of the present invention is to provide a character recognition method that can reduce the rejection rate and improve the recognition rate.

[0004]

In a character recognition device for recognizing an input character, an input character and a candidate having a numerical value corresponding to the input character and showing a similarity between the input character are identified. It is characterized by including a step of matching and a step of judging a candidate such that if the numerical value of the matched candidate does not have a predetermined numerical value, it is rejected, and if it has a predetermined numerical value, it is not rejected.

[0005]

In the character recognition device for recognizing an input character, the input character is matched with a candidate having a numerical value indicating the similarity between the input character and the input character, and the matched candidate If the numerical value of does not have a predetermined numerical value, it is rejected, and if it has a predetermined numerical value, the candidate is judged not to be rejected.Therefore, the rejection rate can be kept low and the recognition rate can be kept high. It can be done efficiently.

[0006]

EXAMPLE In the example of the present invention, a table is created in advance from a distribution of erroneous recognition (first-rank similarity and difference between first-rank and second-rank similarity) by a recognition experiment or the like. By referring to this table, the rejection is performed according to the similarity distribution of erroneous recognition to be rejected. That is, in the present embodiment, in the recognition device using the degree of similarity or the distance value, the degree of similarity (or the distance value etc.) of the first-ranked candidate and the first-ranked candidate when misrecognized in advance by the recognition experiment. The distribution of the difference in the degree of similarity (or distance value) with the second-ranked candidate is made into a table, and the reject judgment is a reject method in which the judgment is made using the table reflecting the distribution of this misrecognition. .

This embodiment will be described in detail below.

FIG. 1 is a flow chart for explaining an embodiment of the present invention, FIG. 2 is a block diagram of a character recognition device used for implementing the embodiment of the present invention, and FIG. 3 is used for reject judgment. FIG. 4 is a diagram showing a first table which is used for reject judgment, FIG. 4 is a diagram showing a second table used for reject judgment, and FIG. 5 is a diagram showing a third table used for reject judgment. .

An outline of the character recognition device shown in FIG. 2 will be described.

In FIG. 2, 1 is a 1-character input unit for inputting 1 character as image information into the apparatus, 2 is a feature conversion unit for converting the input image information into a matching feature, and 3 is an input feature. A matching unit that matches the standard pattern of characters and outputs the top two candidates, 4 is the recognition result 1
It is a reject determination unit that determines whether the position candidate is suspicious. The reject determination unit 4 is composed of a reject table indicating a creating method and a determination unit using the reject table. Matching is based on similarity,
The degree of similarity is between 0 and 10,000.

First, a recognition experiment is performed on all the recognition target characters of the character recognition device, and if all the recognition target characters are equal to or more than this value in the 1st place candidate similarity, it is a sufficiently large value that is reliable as a correct answer. Select one. here,
It is set to 9,000. Next, if the first-ranked candidate similarity is less than this value, one small value that may be determined to be rejected is selected. Here, it is set to 6,000.

Therefore, when the 1st candidate similarity is 9,000 or more, the result of the reject determination is correct, and when the 1st candidate similarity is less than 6,000, the reject determination result is rejected. And

First place candidate similarity is 6,000 or more and 9000
If it is less than 0, reject judgment is performed using the table showing the method of creation below.

At the time of creating a table, in order to reduce the capacity of the table, the element number corresponding to the first-rank similarity is determined by subtracting 6,000 from the first-rank candidate similarity and dividing by 50. This results in Table 1.

[0015]

[Table 1]

As a result, the size of the table is set to 3,000.
0 (9,000-6,000) to 60 (3,000
/ 50), and the reject table is composed of as many characters as the recognition target characters of the table of size 60.

Next, a method of creating a table for each character, for example, the character "A" will be described. 1st place candidate character is "A" and 1st place similarity is 6,000 or more and 9,000
When it is less than the above, the rejection judgment is performed by the table for "A".

As a result of the recognition experiment, it is assumed that the result of misrecognition (incorrect answer) is shown in Table 2 among the candidates for which the first candidate is "A".

[0019]

[Table 2]

Based on this result, the table Table_A [60] when the first-ranked candidate is "A" is created as shown in Tables 3, 4, 5, and 6 below.

(1) If the 1st place similarity is less than 6,000 or more than 9,000, it is not reflected in the table creation. As a result, Table 3 is obtained.

[0022]

[Table 3]

(2) The maximum value of the difference between the similarity of the first candidate and the similarity of the secondary candidate for each i of the element number i (0 ≦ i <60) corresponding to the first candidate similarity. And the maximum value is set as the value of Table_A [i]. When there is no maximum value, it is set to 0. Therefore, it becomes Table 4 and the other T
The value of available_A becomes 0 (see FIG. 3).

[0024]

[Table 4]

(3) For each of the element numbers 0 to 59 corresponding to the 1st place candidate similarity, the value of the table (Table 4) defined in (2) is set to Table_A [i] if i <j. ≧ T
It is set so that it is possible to satisfy the following: able_A [j], (0 ≦ i, j <60) This is because the smaller the 1st place candidate similarity is, the lower the reliability is (the probability that the answer is correct is low), and therefore the difference in the similarity between the 1st place and the 2nd place will have to be large. Therefore, Table 5 is obtained (see FIG. 4).

[0026]

[Table 5]

(4) In order to improve the reliability of the reject (to reduce the error rate), an appropriate value is added to each value of Table_A. Here, 50 is added. Therefore, Table 6 is obtained (see FIG. 5).

[0028]

[Table 6]

The above table creating operation is performed for all the recognition target characters of the character recognition device.

Rejection judgment will be described based on the flowchart of FIG.

In step S1, the first candidate character code
Enter the first-ranked candidate character similarity (rui1) and the second-ranked candidate character similarity (rui2).

In step S2, it is determined whether the first-ranked candidate similarity is 9,000 or more. If the 1st place candidate similarity is 9000 or more, the process proceeds to step S6, and if it is less than 9,000, the process proceeds to step S3.

In step S3, it is determined whether or not the first-ranked candidate similarity is less than 6,000. If the first-ranked candidate similarity is less than 6,000, the process proceeds to step S5, and if it is not less than 6,000, the process proceeds to step S4.

In step S4, the table corresponding to the first rank candidate character is set to Table [60]. The element number corresponding to the first-ranked candidate similarity is num (num
= (1st place similarity-6,000) / 50). Step S7
Proceed to.

In step S5, reject (reject)
And go to the end.

In step S6, without rejecting,
Proceed to the end.

In step S7, it is determined whether the first-order similarity degree-the second-order similarity degree> Table [num]. 1
Degree similarity-second degree similarity> Table [num],
Go to step S8. If not, step S9
Proceed to.

In step S8, without rejecting,
Proceed to the end.

At step S9, the process is rejected and the process goes to the end.

An example of actual rejection judgment is shown below.

Table 7 shows the recognition result of four characters.

[0042]

[Table 7]

The first character has a first-rank similarity of 9,500,
Since it is 9,000 or more, it is not rejected by the reject judgment.

The second character has a first-rank similarity of 5,500,
Since it is less than 6,000, it is rejected by the reject judgment.

The third character and the fourth character each have a first-rank similarity of 6,000 or more and less than 9,000, and the first-rank candidate is "A". Rejection is determined by Table_A) created as described above.

For the third character, the element number corresponding to the first similarity of the third character is 40 (= (8,000-6,
000) / 50) and 8,000-7,500 = 5
00> 350 = Table_A [40] is satisfied, and therefore it is not rejected.

For the fourth character, the element number corresponding to the first similarity of the fourth character is 20 (= (7,000−6,
000) / 50) and 7,000-6,500 = 5
Since 00 ≦ 850 = Table_A [20], it is rejected (rejected).

[0048]

EFFECT OF THE INVENTION In a character recognition device for recognizing an input character, matching is performed by matching an input character with a candidate having a numerical value corresponding to the input character and indicating a similarity with the input character. If the candidate has a numerical value that does not have a predetermined numerical value, the candidate is rejected, and if it has a predetermined numerical value, the candidate is judged not to be rejected.Therefore, the rejection rate can be kept low and the recognition rate can be kept high. The recognition can be performed efficiently.

[Brief description of drawings]

FIG. 1 is a flowchart illustrating an embodiment of the present invention.

FIG. 2 is a block diagram of a character recognition device used to implement an embodiment of the present invention.

FIG. 3 is a diagram showing a first table used for reject determination.

FIG. 4 is a diagram showing a second table used for reject determination.

FIG. 5 is a diagram showing a third table used for reject determination.

[Explanation of symbols]

 1 1 Character input unit 2 Feature conversion unit 3 Matching unit 4 Rejection judgment unit

Claims (1)

[Claims] 1. A character recognizing device for recognizing an input character, the step of matching an input character with a candidate having a numerical value corresponding to the input character and having a degree of similarity with the input character. A character recognition method, which comprises: rejecting if the numerical value of the matched candidate does not have a predetermined numerical value, and rejecting if the numerical value does not have a predetermined numerical value.