JPH0520492A - Document recognizing/correcting device - Google Patents

Abstract

PURPOSE:To obtain a document recognizing/correcting device which can effectively correct the documents by giving the automatic correction equal to that done by an operator to the same character, at the same time giving no wrong automatic correction to the recognized correct characters, and correcting automatically and accurately only the misrecognized characters. CONSTITUTION:When the character images are recognized by a character recognizing means 100, a 1st correction means 200 corrects the wrong characters through an operator. Then a retrieving means 300 retrieves a character string of a prescribed number of characters including those corrected by the means 200 out of the characters recognized by the means 100. A 2nd correction means 400 corrects the character equal to the misrecognized one corrected by the means 200 and included in the retrieved character string into the character corrected by the means 200.

Description

Detailed Description of the Invention

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document recognition / correction device for automatically correcting a document recognition device for recognizing a paper document.

[0002]

2. Description of the Related Art Research has been conducted on a document recognition device for recognizing characters and figures printed on a paper document and inputting them to a document editing device such as a word processor. However, it is very difficult to get 100% recognition rate for character recognition,
In the current situation, it is necessary for the operator to confirm the recognition result and correct any erroneously recognized characters (Image Processing Handbook Shokoido 20.3 Character Recognition Device (OCR) p. 4).
82-490).

[0003]

Since this work must be performed for all the recognized results, a great burden is imposed on the operator. Further, in character recognition, since the target document image is input from the same image input device under the same conditions, the same erroneous recognition is likely to occur for the same character. By using this, it is easy to think that if the same character as the character corrected by the operator is automatically corrected with the same correction made by the operator, the burden on the operator will be reduced. To be However, if this automatic correction is performed by comparing only one character of the recognition result, the correctly recognized character is also corrected, which has a problem of having an adverse effect.

The present invention has been made in view of the above points, and it is an object of the present invention to perform the correction by the operator in order to reduce the burden of the correction when the recognition result is corrected. Efficient correction by automatically making the same correction as the correction for the same character, but not automatically making mistaken automatic correction for correctly recognized characters It is to provide a document recognition correction apparatus and method capable of performing the following.

[0005]

In order to solve the above-mentioned problems, the document recognition / correction device of the present invention is, as shown in FIG. 1, a character recognition means 100 for recognizing a character image and a character recognition means 100 for recognizing the character image. And a character recognizing means 10 for correcting the recognized recognition result according to an operator's instruction.
The first correction means 20 from the characters recognized by 0
Searching means 300 for searching a character string having a predetermined number of characters including the character corrected by 0, and this searching means 300
Second correction means 400 for correcting the same character as the misrecognized character corrected by the first correction means 200 in the character string searched for by the corrected character corrected by the first correction means 200. It has and.

Further, the document recognition / correction apparatus of the present invention uses the above-mentioned document recognition / correction apparatus, and for a character positioned before the position of the character corrected by the first correction means 200. When the character is corrected by the second correction means 400, the operator corrects the character with confirmation.

[0007]

When the character recognizing means 100 recognizes the character image, not all the characters are recognized correctly, so the operator first corrects the erroneous character by the first correcting means 200. . Next, the character recognition means 10
The first correction means 20 from the characters recognized by 0
A character string having a predetermined number of characters including the character corrected by 0 is searched by the searching means 300, and the second correcting means 400 is the first correcting means 20 in the searched character string.
The same character as the misrecognized character corrected by 0 is corrected to the corrected character corrected by the first correction means 200. By doing this, since the search for the same misrecognized character is performed using the character string that includes the preceding and succeeding characters, even if only one character is the same, it is correctly recognized and correction is made for unnecessary characters. Is not performed, and only misrecognized characters can be corrected accurately.

Further, for the character positioned before the position of the character corrected by the first correction means 200, the second
When the character is corrected by the correction means 400, the character is corrected with the confirmation of the operator, so that it is possible to prevent the correction of the corrected portion from being an error.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 2 shows an outline of the entire document recognition apparatus. An image input section 1, an image memory 2, a character image extraction section 3, a character recognition section (OCR) 4, a recognition result storage memory 5,
Correction unit 6, storage unit 7, document file storage device 8, control /
It comprises an operation unit 9. A paper document is digitally input from the image input unit 1 and its original image is stored in the image memory 2. The input document image is displayed on the display device 91 such as a CRT via the character image extraction unit 3 and the control / operation unit 9. The operator extracts only the character image area by using the pointing device 93 such as a mouse while looking at the original image. FIG. 3A shows a state in which only the character image area is designated for the actual document 31. The area enclosed by the dotted rectangle is designated by the operator. The character image area information thus designated is stored in the table 32 as shown in FIG. The first and second columns of the table are the upper left coordinates of the character image rectangular area, and the third and fourth columns are the width and height of the character image rectangular area. As a method for extracting this character image area, in addition to the above-mentioned operator-extracting method, Japanese Patent Laid-Open No. 2-159690
It is also possible to use a method of automatically separating a character and a figure by extracting the characteristics of a black pixel block as shown in Japanese Patent Publication No. The character recognition unit (OCR) 4 recognizes the designated character image area by using the image memory 2 and the table 32 showing the character image areas extracted by the character image extraction unit 3. As a result, the recognition result is output to the recognition result storage memory 5 as the recognition result storage table 51 in a table format as shown in FIG. 4 for each character image area.

Next, the operation of the correction unit 6 will be described with reference to the flowchart of FIG. First, the operator corrects the recognition result in the recognition result storage memory 5 until the correction process is completed (until Y in step 611 in FIG. 6) (step 612). This is because with the current character recognition technology, the recognition rate of the character recognition unit 4 does not reach 100%, and confirmation / correction by the operator is necessary. When making a correction, the recognition result stored in the recognition result storage memory 5 is displayed on the display device 91 such as a CRT via the correction unit 6 and the control / operation unit 9, and the operator displays the recognition result. While looking, correction is performed using the pointing device 93 such as a mouse or the keyboard 92. In the result display, the recognition result is displayed at the position where the original character image was, and is displayed in the same state as the original image as much as possible so that the misrecognized character can be easily found. When the operator sees such a display and finds the erroneously recognized character, the operator correction section 61 uses the pointing device 93 to move the cursor 94 onto the erroneously recognized character as shown in FIG. select. Then, using the keyboard 92, the operator inputs correct characters by using kana-kanji conversion or the like. This input character replaces the misrecognized character and is corrected. At this time, "position of corrected character", "corrected character" and "replaced character"
And are recorded (step 612).

Next, in the automatic correction unit 62, after the correction by the operator, the unrecognized character string (from the "position of the corrected character" in the recognition result storage memory 5 to the last character) is recognized. The character string) is searched for two characters including the "corrected character" and one character after the character, and the same character string is searched (step 621). If the same character string is found (Y in step 621), the operator corrects the same character as the "corrected character" (replaces the "corrected character" with the "replaced character"). The same process is performed (step 622). If they are not the same character string (N in step 621) and all unmodified character strings have not been searched (N in step 623), the same unprocessed character string as described above is searched for (step 621). ) And the correction process (step 622) are repeated. If all uncorrected character strings have been searched (Y in step 623), the process proceeds to step 624. At this time, the uncorrected character string is subjected to the same correction as the correction made by the operator.

Thereafter, similar processing is performed on the recognized character string of the already-corrected portion (the character string from the first character in the recognition result storage memory 5 to the "position of the corrected character"). That is, the recognized character string of the corrected part is searched for two characters including the "corrected character" and one character after the character, and the same character string is searched (step 624). If the same character string is found (Y in step 624), a display is displayed asking the operator to confirm whether or not it may be corrected (step 625), and if OK (Y in step 626), "correction" is performed. Corrected by the operator for the same character as the "corrected character"("replacedcharacter" for "corrected character")
The same processing as that of (replaced by) is performed (step 627).
If it is not OK (N in step 626), no correction is performed and the process returns to step 624. If it is not the same character string (N in step 624) and all the corrected character strings have not been searched (N in step 628), the same correction processing as described above is performed for the next corrected character string (step 624). ) And the correction process (steps 625, 626, 627) are repeated. If all the corrected character strings have been searched (Y in step 628), the process returns to step 611 of the operator correction unit 61.

In the correction processing of the already-corrected character string, since the operator has already corrected the portion, confirmation is requested (steps 625 and 626) before automatic correction is performed.

As an example of the processing of the automatic correction unit, FIG. 7 shows how the automatic correction unit 62 corrects the correction by the operator shown in FIG. In FIG. 7A, 2
When the erroneously recognized character "between" on the line is corrected (it is assumed that the operator did not notice the erroneously recognized character "between" on the first line), it becomes as shown in FIG. 7 (b). Next, the processing of the automatic correction unit 62 is performed. In the automatic correction unit 62, a character string "intermediate" including one character after the character to be corrected is added to the uncorrected portion (that is, from the character "between the character corrected by the operator" to the last character). If the same character string is searched for, the same correction as the operator is made. In this case, the same character string "interworking" exists on the third line, so correct it ("interval" → "question")
I do. The fourth line also has the character "ma", but it is not modified because it is different from the character string "between". Next, for the already-corrected part (that is, from the first character to the character before the character "between" corrected by the operator), include the character after the character to be corrected in the same way as the processing for the uncorrected character string. Search for the string "intertitle". In this case, since the same character string "interworking" is present on the first line, the operator is asked whether or not to make a correction, and if it is OK, the correction ("interval" → "question") is performed. In this way, only the characters that are erroneously recognized (“between” on the 1st and 3rd lines) are corrected, and for the same characters that should not be corrected (“between” on the 4th line), No modifications have been made. This correction process utilizes that the same word is often used in the same document or the same page.

By performing the above-mentioned correction processing, it is possible to create an initial target document. The storage unit 7 converts the corrected document into a document format that can be handled by a document editing device such as an existing word processor, and stores the document file in the document file storage device 8.

In addition to the processing in the automatic correction section 61 described in the above embodiment, the following processing can be changed. (1) The automatic correction unit 61 of the above-described embodiment searches for the same character string with two characters including one character after the correction target character. Searching by a character string of two or more characters including each character or multiple preceding and following characters. (2) Although the automatic correction unit 61 of the above-described embodiment performs the search only once, it should perform the search a plurality of times using a plurality of character strings. For example, after searching for the same character string with two characters including the one character after the correction target character, change to search with the two characters including the one character before the correction target character for the second time. There is. (3) In the automatic correction unit 61 of the above-described embodiment, the search character string is a character string that includes characters just before and after the correction target character, but includes this by the character type (kana, kanji, alphanumeric characters, etc.). To change characters adaptively. For example, if the correction target character is a kanji character, the previous character is a kana character, and the subsequent character is a kanji character, the search character string includes the latter character. As a result, there is an effect that a more effective search (a hit rate at the time of searching becomes higher) becomes possible than the embodiment. (4) In the above-described embodiment automatic correction section 61, the operator is requested to confirm when correcting the already-corrected character string, but the correction is automatically performed without doing this. Or
When correcting an uncorrected character string, ask the operator for confirmation. (5) In the above embodiment, the case where the operator corrects only one character is targeted, but a case where a plurality of consecutive characters are modified can be processed with the same configuration.

[0017]

As described above, according to the present invention,
Since the same erroneously recognized character in the recognized character group is corrected according to the correction made by the operator, the burden on the operator at the time of correction can be reduced. Also, since the search for the same misrecognized character is performed using the character string that includes the preceding and following characters, only 1
Even if the characters of the characters are the same, the characters that are recognized correctly and do not need to be corrected are not corrected, and only the erroneously recognized characters can be corrected accurately. By separating the uncorrected part and the already-corrected part, different processing can be performed on the two parts. In other words, when the character string of the uncorrected part is the same as the corrected character string, the correction is automatically performed, but the already-corrected part is the part that the operator has already inspected, so make the correction after confirming it. Be seen. By dividing the processing in this way, it is possible to prevent an erroneous correction of the already-corrected portion.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an outline of the present invention.

FIG. 2 is a block diagram showing an outline of the entire document recognition device.

FIG. 3 shows an example of character image area extraction.

FIG. 4 shows a table of recognition results from a character recognition unit (OCR).

FIG. 5 is an example of correction by an operator correction unit.

FIG. 6 is a flowchart showing an algorithm of a modified portion.

FIG. 7 is an example of correction by a correction unit.

[Explanation of symbols]

1 ... Image input unit, 2 ... Image memory, 3 ... Character image extraction unit, 4 ... Character recognition unit (OCR), 5 ... Recognition result storage memory, 6 ... Correction unit, 7 ... Storage unit , 8 ...
Document file storage device

Claims (2)

[Claims] 1. A character recognizing means for recognizing a character image, a first correcting means for correcting a recognition result recognized by the character recognizing means according to an instruction of an operator, and a character recognizing means for recognizing the character recognized by the character recognizing means. From the first
Searching means for searching a character string having a predetermined number of characters including the character corrected by the correcting means, and the same misidentified character corrected by the first correcting means in the character string searched by the searching means. And a second correcting means for correcting the character of the above to the corrected character corrected by the first correcting means. 2. The document recognition correction apparatus according to claim 1, wherein a character positioned before the position of the character corrected by the first correction unit is converted by the second correction unit. A document recognition / correction device having a confirmation means for correcting a character upon confirmation by an operator. [0001]