JPH05151384A - Correcting method for recognition character - Google Patents

Abstract

PURPOSE:To execute a correction by making the memory capacity for storing correcting information small, and decreasing the number of execution times of similar operations, in a correction processing of a recognized character. CONSTITUTION:By an image input part 1, a character pattern is inputted, and by a character recognizing part 2, a reject code for giving a feature quantity of the pattern, a candidate character and erroneous recognition information is detected, from the reject code, with regard to only the pattern whose possibility of erroneous recognition is high, the feature quantity, and the candidate character and a character pattern coordinate are stored in a feature quantity memory 5v and a reject information memory 9, respectively. The candidate character of the reject information memory 9 is displayed as a correction object on a display part 4, and corrected after an operator's confirmation. Also, a candidate character of the pattern being similar to its corrected candidate character is subjected to similar operation, and extracted and displayed, and it is also corrected after the operator's confirmation. By storing only the information of the pattern whose possibility of erroneous recognition is high, the storage capacity is curtailed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of correcting recognized characters in character recognition.

[0002]

2. Description of the Related Art In recent years, character recognition devices have been developed, but their recognition accuracy is a problem, and many of them have a function of correcting recognized characters.

A method of correcting a recognized character in a conventional character recognition device will be described below with reference to the drawings. FIG. 5 is a block diagram showing the structure of a conventional character recognition device capable of correcting characters. In the figure, the character code, the character pattern, the feature amount, and the reject code of the recognized character output from the character recognition unit 2 are stored in the character code memory 3, the character pattern memory 10, the feature amount memory 5, and the reject information memory 9, respectively. Remember only the number. Next, the recognized character and the character pattern thereof having a high possibility of being erroneously recognized are read from the character code memory 3 and the character pattern memory 10 and presented to the operator on the display unit 4. When the operator inputs a correct character code to correct this character code from the correction information input unit 6, and even when the operator does not instruct the correction, the feature amount of the character pattern in the feature amount memory 5 And the similarity with the feature amount of another character pattern are calculated, the character pattern and the character code having a similarity greater than a predetermined value are displayed to the operator, and the confirmation unit 8 determines whether or not to correct the character code. The operator is asked to confirm, and the correction is made when the operator permits the correction.

[0004]

In such a conventional recognition character correction method, a feature amount and a reject code must be stored in a memory for all characters, so that a large memory capacity is required. However, there is a problem that the processing time is long because the similarity calculation is uniformly performed even for a character having a low possibility of being erroneously recognized.

An object of the present invention is to solve the above problems, and an object thereof is to provide a method of correcting a recognized character which requires only a small storage memory capacity and requires a small number of times of similarity calculation.

[0006]

In order to achieve the above-mentioned object, the present invention reads a character pattern by image input means, extracts the feature quantity of the pattern by character recognition means, and extracts the character from the feature quantity. A code and a reject code are obtained, the character code is stored in the first storage means, and a characteristic amount of a specific character determined to have a high possibility of being misrecognized by the reject code and information required for correction are stored in the second storage. In the character correction process, the specific character of the second storage means is displayed to the operator by the display means, and the operator selects an arbitrary first character of the displayed specific characters from other characters. When you change to the second character,
Characters in the second storage means, which are similar to the first character with a predetermined value or more, are obtained by a similarity calculation and displayed, and the operator confirms the character to display the second character or other characters. This is a method of correcting the recognized character so that the character is corrected to.

[0007]

According to the present invention, in the above configuration, the second storage means stores the characteristic amount of the specific character having a high possibility of being erroneously recognized and the information required for the correction, and displays the specific character to the operator in the character correction process. Then, when the operator changes any first character of the specific characters to another second character, the second character
A character in the storage means that is similar to the first character is sought and displayed, and is corrected to the second character or another character by the operator's selection.

[0008]

【Example】

(Embodiment 1) Hereinafter, a method of correcting a recognized character according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a character recognition apparatus using a method for correcting recognized characters according to an embodiment of the present invention.
In the figure, 1 is an image input unit that photoelectrically converts a document and outputs binarized data to the document pattern memory 11. Reference numeral 2 is preprocessing, feature extraction, and a character code and a reject code from the feature amount for the binarized data. A character recognition unit 3 which performs character recognition processing such as matching to output character pattern coordinates, feature amounts, candidate characters, and reject information in the document pattern memory 11, and 3 is a candidate for storing the candidate characters recognized by the character recognition unit 2. A character memory, 4 is a display unit for displaying the contents of the document pattern memory 11, candidate characters in the candidate character memory 3, a message for confirmation of correction from the confirmation unit 8, and the like, and 5 is a feature amount stored by the character recognition unit A feature amount memory, 6 is a correction information input section for the operator to input correction information, and 7 is a confirmation section 8 which is a matching section when the operator approves the correction. Control unit for modifying the contents of the candidate character memory 3 to control the 0, the reject information memory for storing a reject information output by the character recognition unit 2 9, 1
0 is a matching unit for collating the feature amounts in the feature amount memory 5 with each other, 11 is a document pattern memory for storing the binary data from the character input unit 1, and 12 is a character output from the character recognition unit 2 and is rejected. It is an AND gate for writing only the information regarding the characters written in the reject information memory 9 and the feature amount memory 5. Although the candidate character is stored as a character code and the reject information is stored as a reject code and a feature amount in the embodiment, other means such as a character font may be used.

FIG. 2 is a block diagram showing the configuration of the storage means of the character recognition apparatus using the method for correcting the recognized characters according to the embodiment of the present invention and the peripheral devices thereof. In the figure, reference numeral 21 is an image scanner for reading a document, converting it into bit data and outputting it, and 22 is a RAM, which is a document pattern memory 1 for storing bit data output by the image scanner 21.
1, a candidate character memory 3 for storing a recognized character, a feature amount memory 5 for storing a feature amount of a character having a high possibility of being erroneously recognized among the candidate characters in the candidate character memory 3, and information about a reject. The reject information memory 9 includes a register area 27 used for processing, and a correction information area 28 for storing correction information for an erroneously recognized character from the operator. Reference numeral 23 denotes a ROM, which is a dictionary area 30 that stores a feature amount and a character code corresponding to the ROM, and a program storage area 31 that stores a program for controlling the correction operation.
It has and. Reference numeral 24 denotes a processing circuit that performs processing in accordance with the control program stored in the program storage area 31,
Reference numeral 25 is a keyboard for inputting data, and 4 is a display unit.

The mutual relationship and operation of the above-mentioned components will be described below with reference to the drawings. First, the preceding process for correction will be described. FIG. 3 is a flowchart showing the preceding processing operation for correction. At step s100, i and j of counters for counting the order of operations are set to 1 and initialized. Note that i is the order of characters to be recognized, and j is the storage order of the feature amount memory 5 and the reject information memory 9. The process starts from i = 1, that is, the first character of the candidate character string. Step s
In step 101, the binary data of the document input from the image input unit 1 is subjected to preprocessing such as noise removal and character segmentation, and the process proceeds to step s102. If the i-th character pattern to be recognized in step s102 does not exist, the process proceeds to step s110, and if it exists, step s10.
In step 3, the character recognition process extracts the characteristic amount e [i] of the character pattern and the candidate character corresponding to the characteristic amount, and then proceeds to step s104. In step s104, the candidate character code a [i] is stored in the candidate character memory 3, the process proceeds to step 105, and the possibility of erroneous recognition is determined based on the reject code of the recognized character pattern. Step s1 as a character pattern to be rejected
If the process proceeds to 06 and does not reject, step s1
Move to 09. It should be noted that the erroneous recognition possibility determination in step s105 may be performed by using a unit that makes it easy to erroneously recognize a predetermined specific character pattern.

In step s106, when the i-th character pattern is the j-th character pattern to be rejected,
The feature amount e [i] is stored in the feature amount memory 5, and in step s107, the address storing a [i] in the candidate character memory 3, the address storing e [j] in the feature memory 5, and the document pattern. The coordinates of the character pattern in the memory 11 and b of the reject information memory 9 are
Store in the [j], c [j], and d [j] areas. Next, the process proceeds to step s108, 1 is added to j, the process proceeds to step s109, 1 is added to the value of i, the process returns to step 102, and the process for recognizing the next (i + 1) th character pattern is performed. By sequentially repeating the above processing until the last character pattern, the number of pieces of character pattern information that are likely to be erroneously recognized are stored in the reject information memory 9. When the processing is completed for all the character patterns, the process proceeds to step s110, and 0 is stored in b [j] to indicate the end of the reject information. Note that b [j] = 1 is set for candidate characters that have been corrected in the correction process described below, and therefore,
The addresses of the candidate characters in the candidate character memory 3 are 0 and 1.
Have an address other than. The display unit 4 displays the content of the document pattern memory 11 and the character code in the candidate character memory 3 or the character font corresponding thereto.

Upon completion of the preceding processing, the next correction processing is started. The correction process will be described below with reference to the drawings. FIG. 4 is a flowchart showing the operation of the correction process. The correction process is executed according to the storage order j based on the information in the reject information memory 9. First, in step s10, the counter j is set to 1 and initialized.
In step s11, the contents of b [j] stored in the reject information memory 9 are referred to. If b [j] = 0, there is no candidate character to be modified, so the process ends. b
If [j] ≠ 0, the process proceeds to step s12, and b
It is determined whether the candidate character stored in the candidate memory 3 at the address given by [j] has already been corrected.
That is, as described above, if b [j] = 1, it is determined as a corrected character, so the process proceeds to step s25 and [j] ≠ 1.
If so, the process moves to step s13, the character pattern of the coordinates given by d [j] of the reject information memory 9 to the operator is displayed on the display unit 4, and the process proceeds to step 14 to judge the operator's correction. Ask for. If the displayed character is to be modified to another character, the process proceeds to step s15, and if not, the process proceeds to step s25. In the case of correction, the process proceeds to step s15, the candidate character stored in the candidate memory 3 at the address given by b [j] is replaced with the character input by the operator, and the correction is now performed by the processing of step 16 and subsequent steps described below. A candidate character similar to the original candidate character thus obtained is searched from unprocessed candidate characters, and correction processing is performed by confirmation of the operator. If not corrected in step s14, the process proceeds to step s25, 1 is added to j, and the process returns to step s11 to proceed to the next j + 1-th reject information process.

When the candidate character of the address b [j] is corrected in step s15, the address b following b [j]
Of the uncorrected candidate characters [j + 1], b [j + 2], ..., Candidate characters similar to the character at the address b [j] are searched for by a similar operation, and the process moves to a process for correction by confirmation by the operator. To do. Therefore, first, in step s16, the counter k
Substituting j + 1 for the feature quantity e stored in the feature quantity memory 5 at the address k (= j + 1) in step s19.
The feature amount e of the original character pattern in which [k] is now corrected
[J] is compared and calculated. Before that, it is checked in step s17 whether the reject information corresponding to k remains in the reject information memory 9, and in step s18 it is checked whether the character corresponding to k has been corrected. If not, the process proceeds to step s25. If it has been corrected, the process proceeds to step s23.

In step s19, the reject information memory 9
Of the feature quantity e [j] of the jth character and the feature quantity e [k] of the kth character stored in the feature quantity memory 5 at the address given by b [j] and b [k] in The degree is calculated, and it is determined whether the degree of similarity is larger than a predetermined threshold value. If it is larger, the process moves to step s20, the candidate character is presented to the operator, and if it is desired to correct it, it is corrected in step s22, 1 is set in the address b [k] as corrected in step s24, and in step s23. 1 is added to k, and the characteristic amount of the character at the next address b [j + 2] is compared with the characteristic amount at the address b [j]. In this way, for the corrected character at the address b [j], the address b [j]
Characters such as +1], b [j + 2], ... Are compared for similarity, and similar characters are displayed and corrected by the operator's confirmation.

Below, the 7-character string "Akaasao"
The operation will be specifically described with reference to FIG. In the following description, it is assumed that the character patterns “a” and “o” are patterns that are likely to be erroneously recognized because their feature amounts are similar, and are erroneously recognized as “okai osasa”. I will explain as things.

The image input unit 1 stores the character string pattern to be recognized as binary image data in the document pattern memory 11. Next, in step s101, preprocessing such as noise removal and character cutout is performed on the image stored in the document pattern memory 11, and 1 is applied to the cut out character pattern.
The preceding process after step s102 is sequentially executed from the th character pattern. As a result, the candidate character is stored in the candidate character memory 3. Table 1 shows the configuration of the candidate character memory 3 in a table.

[0018]

[Table 1]

In Table 1, the number i is the character sequence of the character string to be recognized, and a [i] is the content of the candidate character, that is, the code of the candidate character. In the case of the embodiment, the character pattern "a" is erroneously recognized as "o" and stored.

For a character pattern which is likely to be erroneously recognized, the characteristic amount is stored in the characteristic amount memory 5 in step s104. Table 2 shows the configuration of the feature amount memory 5 in a table.

[0021]

[Table 2]

In Table 2, the number j means the jth character pattern rejected as a character pattern with a high possibility of being erroneously recognized, and its feature amount e [j] is j in the memory.
Remembered th. In the case of the embodiment, the character pattern "a"
Is assumed to have a high possibility of being erroneously recognized, and e [1], e [2], e
It is stored as [3].

Further, (Table 3) is a table showing the structure of the reject information memory.

[0024]

[Table 3]

In Table 3, the number j means that the rejected character pattern is the jth character pattern and also corresponds to the jth character in the reject information memory. & A [j] is the address of the rejected character code a [i] in the candidate character memory 3, & c [i] is the address of the rejected character's feature amount e [j] in the feature amount memory 5, d
[J] is the character pattern memory 11 of the rejected character
Is the coordinate at. When j = 4, the address value 0 is stored in b [4] to indicate that there is no reject information thereafter.

The correction process is performed in order from j = 1 of the reject information memory shown in (Table 3). When the candidate character stored in the candidate character memory 3 is displayed at the address b [1] value (= & a [1]) given by j = 1, "O" is displayed. Fix it. At this time, the value c
Each time a candidate character having a feature quantity similar to the feature quantity of an address given by [1] (= & e [1]) is detected and displayed by a similar operation in j = 2 and j = 3, each time it is detected, Correct at the operator's discretion. In the case of the embodiment, when the first "A" of the character string is corrected to "O", the fourth "O" is displayed, and it is corrected to "A" and then "7th".
“O” is displayed and remains “O” without correction. Next, since j = 4 and the address value b [4] is 0, there is no reject information and the correction process ends.

As described above, according to the recognized character correction method of the embodiment of the present invention, the memory capacity is stored by storing the characteristic amount and the information necessary for the correction only for the character having a high possibility of erroneous recognition. Is small, and the number of times of similarity calculation is small, so that it is possible to realize recognition character correction processing with higher cost performance than in the past.

In the present embodiment, character recognition has been described as an example, but it goes without saying that it can be applied to the case of voice recognition.

[0029]

As is apparent from the above-described embodiments, the present invention reads a character pattern by the image input means, extracts the characteristic amount of the pattern by the character recognition means, and extracts the character code and the reject from the characteristic amount. A code is obtained, the character code is stored in the first storage means, and the characteristic amount of the specific character determined to have a high possibility of being erroneously recognized by the reject code and the information required for correction are secondly stored.
In the character correction process, the second
The specific character of the storage means of the above is displayed to the operator by the display means, and the operator selects any first of the displayed specific characters.
When the character of is modified to another second character, the character in the second storage means which is similar to the first character by a predetermined value or more is obtained and displayed by a similarity calculation, and the character is displayed. Is confirmed by the operator and corrected to the second character or another character, thereby extracting from the original character pattern only the character with a high possibility of being erroneously recognized. Since the stored feature amount and the information necessary for correction are stored, the number of target characters for matching calculation is smaller than in the past, the memory capacity and the amount of matching calculation are less, and the correction process of recognized characters with high cost performance is performed. realizable.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a character recognition device using a method for correcting a recognized character according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a storage unit and its peripheral device in a character recognition device using a method for correcting a recognized character according to an embodiment of the present invention.

FIG. 3 is a flowchart showing an operation of a preceding process in the recognition character correction method according to the embodiment of the present invention.

FIG. 4 is a flowchart showing the operation of a character correction process in the recognized character correction method according to the embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a character recognition device using a conventional method for correcting recognized characters.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 image input section (image input means) 2 character recognition section (character recognition means) 3 candidate character memory (first storage means) 4 display section (display means) 5 feature amount memory (second storage means) 9 reject information Memory (second storage means)

Claims (1)

[Claims] 1. A character pattern is read by an image input unit, a feature amount of the pattern is extracted by a character recognition unit, a character code and a reject code are obtained from the feature amount, and the character code is stored in a first storage unit. Stored in the second storage means, the characteristic amount of the specific character determined to have a high possibility of being erroneously recognized by the reject code and the information required for correction are stored in the second storage means, and in the character correction processing, the second storage means is stored. The specific character is displayed to the operator by the display means, and the operator corrects an arbitrary first character of the displayed specific characters to another second character,
Characters in the second storage means, which are similar to the first character with a predetermined value or more, are obtained by a similarity calculation and displayed, and the operator confirms the character to display the second character or other characters. How to correct the recognized character so that it will be corrected to the character.