JPH06195521A - Character recognizing method - Google Patents

Abstract

PURPOSE:To accurately recognize a Japanese document by judging that the Japanese document is not the result of recognition when the number of candidate words obtained by making a search only as to a 1st candidate character in one line is less than a threshold value. CONSTITUTION:A language processing is performed for the character recognition result of a character recognition part 13 which recognizes and converts characters of character image data, obtained by a character segmentation part 12 segmenting image data read by an image reader into character image data, character by character, into character codes. At this time, a post processing necessity/nonnecessity decision part 15 judges that the recognition result is not the result of the recognition of the Japanese document if the number of characters in one line exceeds the threshold value, there are successive characters of single character kind more than the threshold value among 1st candidate characters in one line, and candidate words less than the threshold value are obtained as a result of a search for only 1st candidate words in one line, so that the language processing is not performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is designed to print characters, dot characters and handwritten character patterns of newspapers, magazines, novels, etc. according to JIS.
The present invention relates to a character recognition method for converting into code information such as a code.

[0002]

2. Description of the Related Art The original text of a document to be recognized is shown in (Table 1).

[0003]

[Table 1]

The recognition result of the original sentence shown in (Table 1) is shown in (Table 2).
Shown in.

[0005]

[Table 2]

When post-processing using language processing is performed on the recognition result, the post-processing detects the combination of characters that form a Japanese word by combining the characters in the candidate character group and adopts that word as the correct character. To do. Therefore, in the post-processing, the 0th row and 1st column'position 'and the 1st row and 2nd column'phase' form the word'phase ', so the correct character'transition' is converted to the incorrect character'phase '. However, there is a problem that the recognition rate decreases.

(Table 3) shows the recognition results that are adversely affected.

[0008]

[Table 3]

[0009]

In the conventional character recognition method, since the post-processing using the language processing is performed on all recognition results, the above problem occurs in a document in which characters of a single character type are listed. Occurs. Even if the user specifies the necessity of post-processing, it is not possible to completely prevent the above-mentioned problems from occurring due to a specification error.

In view of this point, the present invention has an object to provide a character recognition method capable of accurately recognizing a Japanese sentence.

[0011]

A character recognizing unit for recognizing character image data cut out by a character slicing unit for slicing image data read from an image reading device into character image data for each character and converting the character image data into a character code. When performing language processing on the character recognition result recognized by, the number of characters in one line exceeds the threshold in the recognition result, and the single candidate character among the first candidate characters in one line is detected. When the number of character types continues for more than the threshold value and the candidate words are searched for only the first candidate character in one line, only the number of candidate words less than the threshold value is obtained. It is determined that the recognition result is not the result of recognizing a Japanese sentence, and the language processing is not performed to prevent the recognition rate and the recognition speed from decreasing.

[0012]

The present invention can correctly recognize a Japanese sentence by the above configuration.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A character recognition method according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a character recognition apparatus using a character recognition method according to an embodiment of the present invention.

Reference numeral 11 is an image reading unit for photoelectrically converting a recognition target document, 12 is a character cutout unit for cutting out image data in character units from image data of the recognition target document, and 13 is character recognition by recognizing image data in a character area. A character recognizing unit for converting into a code, 14 detects a line feed code from a candidate character group, and a line dividing unit for obtaining a recognition result for each line, 15 determines whether post-processing using language processing is necessary for a line A post-processing necessity determining unit for performing, a post-processing unit for performing post-processing using language processing, an output unit for outputting a final recognition result, and a word dictionary.

The operation of the character recognition method according to this embodiment having the above-described structure will be described below with reference to the flow charts of FIGS.

First, in step s21, the image reading section 11 photoelectrically converts the document to be recognized to obtain image data.

Next, in step s22, the character cutout unit 12 cuts out image data in character units from the image data.

Next, in step s23, the character recognition unit 13 performs character recognition based on the character image data. The recognition results are shown in (Table 4).

[0020]

[Table 4]

Next, line division is performed. In step s31, i represents a character position, and -1 is set as an initial value. Then, in step s32, i + 1 = 0 is changed to h
Substitute in ead. The head represents the start position of the line.
In step s33, i + 1 = 0 is substituted for i and the character position is advanced by one.

In step s34, it is checked whether i exceeds the number of recognized characters. If it does not exceed, the process proceeds to step s35 to check whether the first candidate character at the (i = 0) position is a line feed code. If it exceeds, the process proceeds to step s37 and the recognition result is output. If it is not a line feed code, the process proceeds to step s33.

Similarly, when i is advanced one by one and i is advanced to 6, the first candidate character is a line feed code, so i is substituted for tail in step s36. tail
Represents the last character position in the line.

With the above operation, the recognition result range for one line can be obtained. In step s40, the number of characters in the line is calculated.
Then, in step s41, the length is set to (tai
l-head) is substituted.

Then, it is checked whether the length is less than thr1. thr1 is a threshold value that represents the number of characters per line, and is one of the post-processing necessity determination elements. Here, it is assumed that 4 is set.

In step s42, head is substituted for j as an initial value. j is 0. j represents a character position.
Then, in step s43, the first candidate character type at the j position is set to kind. The character type is Kanji. In step s44, the initial value 1 is set in cnt. cnt is used to count the number of consecutive characters of the same character type. In step s45, 1 is added to j.

In step s46, it is checked whether j does not exceed tail. If not exceeded,
In step s47, the first candidate character type at the position (j = 1) is compared with the kind. Since both are Kanji, if they match, the process moves to step s48 and 1 is added to cnt. cnt =
When it reaches 2, cnt is t in step s49.
Check if hr2 is not exceeded. thr2 is the allowable number of consecutive appearances of the same character type, and if this value is exceeded, post-processing will not be performed. Here, 5 is set.

Since (cnt = 2) does not exceed (thr2 = 5), the process proceeds to step s45.

Similarly, when j is advanced one by one and j is advanced to 5, cnt exceeds thr2.

Head is substituted for key (s60).
The key becomes 0. The key represents a candidate word matching start point. In step s61, the word starting with the character at the key position is read from the word dictionary 18.

Since the key = 0 position is "A", reading a word starting with "A" from the dictionary yields (Table 5).

[0032]

[Table 5]

The words in (Table 5) are compared with the candidate characters (s
62). No matching word was obtained, so no counting is performed.

It is checked whether the key matches tail (s64). It does not match. Similarly, the key was advanced one by one, and candidate words were extracted up to tail, but no candidate word was obtained.

The number of coincidences and thr3 are compared (s6
6). thr3 is a threshold representing the number of candidate words,
Here, 3 is set. The number of candidate words is 0.
Therefore, no post-processing is performed on the Bank.

The next line division is performed. In step s32, i + 1 is substituted for head. head = 7.

In step s33, i + 1 is substituted for i and the character position is advanced by one. i becomes 8.

In step s34, it is checked whether i exceeds the number of recognized characters. Since it has not exceeded, the process proceeds to step s35 to check whether the first candidate character at the i position is a line feed code. Since it is not a line feed code, the process moves to s33.

Similarly, i is advanced one by one, and i is increased to 13
Since the first candidate character is a line feed code when i is advanced to, i is substituted for tail in step s36.

With the above operation, the recognition result range for one line can be obtained. In step s40, the number of characters in the line is calculated.
Then, in step s41, the length is set to (tai
l-head) is substituted.

Then, it is checked whether the length is less than thr1. Since length is more than thr1, step s42
Move to.

In step s42, head is substituted for j as an initial value. j is 7. And step s43
In, the first candidate character type at position j is set to kind. The character type is Kanji.

At step s44, the initial value 1 is set to cnt.
To set. In step s45, 1 is added to j. j is 8.

In step s46, it is checked whether j does not exceed tail. Since it has not exceeded, the process proceeds to step s47.

In step s47, the first candidate character type at the j position is compared with the kind. Since both are Kanji, the process moves to step s48 and 1 is added to cnt.

In step s49, cnt is thr2.
Since it does not exceed, the process proceeds to step s45.

At step s45, 1 is added to j. j is 9. In step s46, it is checked whether j does not exceed tail. Since it has not exceeded, the process proceeds to step s47.

In step s47, the first candidate character type at the j position is compared with the kind. Since both do not match, the process proceeds to step s50.

In step s50, the first candidate character type at the j position is substituted for kind. In step s51, the initial value 1 is substituted for cnt.

At step s45, 1 is added to j. j is 10. In step s46, it is checked whether j does not exceed tail. Since it has not exceeded, the process proceeds to step s47.

In step s47, the first candidate character type at the j position is compared with kind. Since both do not match, the process proceeds to step s50.

In step s50, the first candidate character type at the j position is substituted for kind. In step s51, the initial value 1 is substituted for cnt.

At step s45, 1 is added to j. j is 11. In step s46, it is checked whether j does not exceed tail. Since it has not exceeded, the process proceeds to step s47.

In step s47, the first candidate character type at the j position is compared with the kind. Since both do not match, the process proceeds to step s50.

In step s50, the first candidate character type at the j position is substituted for kind. In step s51, the initial value 1 is substituted for cnt.

At step s45, 1 is added to j. j is 12. In step s46, it is checked whether j does not exceed tail. Since it has not exceeded, the process proceeds to step s47.

In step s47, the first candidate character type at the j position is compared with kind. Since both do not match, the process proceeds to step s50.

In step s50, the first candidate character type at the j position is substituted for kind. In step s51, the initial value 1 is substituted for cnt.

At step s45, 1 is added to j. j is 13. In step s46, it is checked whether j does not exceed tail. Since it has not exceeded, the process proceeds to step s47.

In step s47, the first candidate character type at the j position is compared with the kind. Since both do not match, the process proceeds to step s50.

In step s50, the first candidate character type at the j position is substituted for kind. In step s51, the initial value 1 is substituted for cnt.

At step s45, 1 is added to j. j is 14. In step s46, it is checked whether j does not exceed tail. Since it has exceeded, the process proceeds to step s70. That is, in the range of 7 to 13 character positions, the same character type string exceeding str1 does not exist. Therefore, the bank is judged to be a Japanese sentence, and the post-processing unit 1
The process of 6 is performed.

Therefore, the bank is judged to be a Japanese sentence, and the post-processing unit 16 performs the processing.

[0064]

According to the present invention, with this configuration, Japanese sentences can be correctly recognized.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing a control procedure of a character recognition device using a character recognition method according to an embodiment of the present invention.

FIG. 3 is a flowchart showing a control procedure of a character recognition device using a character recognition method according to an embodiment of the present invention.

FIG. 4 is a flowchart showing a control procedure of a character recognition device using a character recognition method according to an embodiment of the present invention.

FIG. 5 is a flowchart showing a control procedure of a character recognition device using a character recognition method according to an embodiment of the present invention.

FIG. 6 is a flowchart showing a control procedure of the character recognition device using the character recognition method in the embodiment of the present invention.

[Explanation of symbols]

 11 image reading unit 12 character cutout unit 13 character recognition unit 14 line division unit 15 post-processing necessity determination unit 16 post-processing unit 17 output unit 18 word dictionary

Claims (1)

[Claims] 1. A character recognition unit for recognizing character image data cut out by a character slicing unit for slicing image data read from an image reading device into character image data for each character and converting it into a character code. When performing post-processing using language processing on the character recognition result, the number of characters in one line is greater than or equal to the threshold in the recognition result, and only one of the first candidate characters in one line is detected. The number of one character type continuously existed more than the threshold value, and as a result of searching the candidate words only for the first candidate character in one line, only the number of candidate words less than the threshold value was obtained. In this case, the recognition result is determined not to be the result of recognizing a Japanese sentence, and no language processing is performed, which is a character recognition method.