JP2002366893A - Document recognizing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a document recognizing method which can improve the recognition precision. SOLUTION: The document recognizing method which limits a document kind for an inputted document image (104), recognizes characters of the document image with the limited document kind (104), and outputs the recognition result to a host device (108) is characterized in that when the character recognition of a recognition request area in the document image is performed, a character string equivalent to a character string entered into the recognition request area which is detected in an area other than the recognition request area and the recognition result are merged and outputted to the host device.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a form recognition method, and more particularly to a form recognition method for recognizing a form having a plurality of equivalent contents, such as a continuous form.

[0002]

2. Description of the Related Art A conventional form recognition method is disclosed in
As described in Japanese Patent Laid-Open No. 25012/1992, the form type is limited to one for the form image input from the image input means, and the character recognition is performed only in the character recognition request area of the limited form type. And outputs the recognition result to the host device. When character types are limited during character recognition, it is necessary to register character types to be used for each form type in advance.

[0003]

Most of the forms currently used include an "item name" (eg, "address") and a character for the "item name" (eg, "Owariasahi-shi, Aichi"). Is written in the right position near the item name or in the lower position.

In the conventional form recognition method, when it is desired to obtain characters to be entered for an "item name" such as an address or an amount of money, the processing time increases if information to be obtained is searched from all areas in the form, and a ruled line and a character Lines are not correctly extracted due to contact such as, and sufficient recognition accuracy cannot be obtained. Therefore, the form type is limited based on the layout characteristics of each form, and the characters to be entered for the "item name" that you want to obtain with the limited form The character recognition process is performed only for a specific area with reference to a form format definition file in which the description area is registered in advance.

[0005] Here, when a plurality of characters for "item name" to be obtained in a form are described in different areas, the conventional form format definition file seems to be the easiest to recognize among them. Had registered one area. However, in this method, there is a case where noise or blur is present in the area registered in the form format definition file in the actual form and characters cannot be recognized if there is no noise or blur in another area where the same contents are described. Occurs. As a means for improving the recognition accuracy when recognizing the character recognition request area of the form format definition file, a method has been proposed in which the character type of the character entered for the "item name" is limited. However, in this method, when creating a form format definition file, the operator needs to register the character type of the character to be entered for the "item name" for each form, which increases the load.

An object of the present invention is to provide a form recognition method capable of improving recognition accuracy.

Another object of the present invention is to provide a form recognition method capable of reducing the work of an operator.

[0008]

According to the present invention, a form type is limited for an input form image, character recognition is performed on the form image with the limited form type, and a recognition result is output to a host device. In the form recognition method, when performing character recognition on a recognition request area in the form image in the limited form type, a character string equivalent to a character string described in the recognition request area is requested by the recognition request. A form recognition method characterized by detecting from an area other than an area, integrating recognition results, and outputting the result to the host device.

[0009] The present invention provides the above-mentioned limited form type,
A form recognition method characterized in that when performing character recognition on a recognition request area in the form image, character types used in character recognition are limited and recognition is performed.

[0010]

FIG. 1 is a block diagram showing an embodiment of a form recognition apparatus to which a form recognition method of the present invention is applied.
In FIG. 1, reference numeral 101 denotes a form recognition unit for recognizing a form;
02 is an image scanner, and 108 is a higher application. The form recognition unit 101 includes a form identification unit 103, a character recognition unit 104, a character recognition result analysis unit 105, a recognition result integration unit 106, and a recognition result output unit 107.

The form identification unit 103 includes the image scanner 1
With respect to the form image of the form input from 02, the form type is determined with reference to the form layout definition file 109 in which the layout characteristics of each form are registered. The character recognition unit 104 performs character recognition on the input character recognition request area using the form format definition file 110 and the character type character dictionary 111, and the character recognition result analysis unit 105 It is determined whether or not to perform character recognition for complementing the recognition result. When performing character recognition for complementation, the character recognition for completion is performed by referring to the common knowledge database 112 and the item name / word collation dictionary 113. Perform processing. The recognition result integrating unit 106 integrates the character recognition result of the character recognition request area registered in the form format definition file 110 with the character recognition result performed for complementing the recognition result, and the recognition result output unit 107 The character recognition result is output to the host application 108. The upper application 108 indicates an apparatus having a function of displaying the input character recognition result on a display, for example.

FIG. 2 shows an example of a form 201 and an example of the form format definition file 110 of FIG. 1 in which the position of information to be acquired of the form is registered. Is registered and is commonly referred to in all the forms.
And an example of a character type character dictionary 111 indicating a dictionary for each character type.

A form 201 includes preprinted item names 205, 206, 207, and 208, and 209, 210, and 21 written in handwritten or printed for each form.
1,212. Form format definition file 1
In an area 10, the area including the item names 205 and 206 of the form and the characters 209 and 210 corresponding to the item names is registered as “character recognition request rectangle upper left position−character recognition request rectangle lower right position”. The common knowledge database 112 sets the relationship between each item name used in an arbitrary form and the character type of the entry character for the item name as “item name 1 = item name 2 = item name 3”.
: Registered with character type A. At this time, it is assumed that the item names connected by "=" indicate equivalence, and if the item names are equivalent, it is considered that the entered characters for the item names are also equivalent. A character type 111 is used to register the character type used as the character to be entered for each item name, separately for a money amount character type dictionary, a number character type dictionary, and an address character type dictionary.

FIG. 3 is a flowchart showing the flow of the processing of the entire form recognition apparatus of FIG. When a form recognition request is generated, a form image is input by the image scanner 102 (step 301). Based on the input form image, the form identification unit 103 determines the form type of the form with reference to the form layout definition file 109 indicating the layout characteristics of each form (step 302). For the character recognition request area registered in the form format definition file 110 corresponding to the determined form type, the character type string dictionary 111 The recognizing unit 104 performs character recognition to obtain “item name” and characters to be entered for “item name” (step 303). The recognition reliability of the entered character for the "item name" calculated by the character recognition unit 104 is calculated (step 304).

A pattern matching method will be described below as an example of a method of calculating the input character recognition certainty factor. To recognize characters, a character dictionary in which patterns (shapes) for each character are registered is used. In character recognition, a character line is extracted by a line extraction process, and the extracted character line is further extracted in character units. The pattern of the black pixels forming the characters in the cut-out character extraction region and the pattern of the black pixels forming the characters registered in the dictionary are superimposed. At the time of superimposition, if the overlapping portion is equal to or more than a predetermined value, the cut-out region is recognized as a character corresponding to the dictionary pattern. At this time, the character certainty factor is calculated based on the amount of the overlapping portion. The certainty is high when the black pixel pattern of the cut-out area and the dictionary pattern are largely overlapped, and the certainty is low when the overlapping portion is small.

In step 304, if the recognition certainty is high, the recognition result integration (step 316) is performed. If the recognition certainty is low, the character recognition result analyzing unit 105 uses the item name / word collation dictionary 113. Then, word matching of "item name" is performed (step 305), and it is determined whether or not "item name" has been obtained (step 306).

When the "item name" is obtained, the character recognition result analysis unit 105 performs a process for complementing the recognition result of the entered character for the "item name". First, the "item name" is searched with reference to the common knowledge database 112, and it is determined whether or not the character type of the character entered for the "item name" is registered (step 307). If the character type of “entry for the item name” is registered, the character recognizing unit 104 re-enters the character type dictionary 1 for the character to be entered for the “item name”.
Then, a character type dictionary corresponding to the character type is selected from 11 and characters to be entered are recognized (step 308).

Then, the recognition certainty of the entered character for the "item name" is calculated (step 309). If the recognition certainty is high, the recognition result is integrated (step 316). From the area other than the character recognition request area registered in the form format definition file 110, the “item name” obtained from the character recognition request area registered in the form format definition file 110
In order to detect a character, the character recognition unit 104 performs character recognition using the entire character type dictionary (no character type limitation) from the character type character string dictionary 111 for all regions in the form (step 310). ).

Originally, the character recognition processing for the entire area in a form takes a long processing time, and line extraction is not performed correctly due to contact between a frame or a line described in the form. This process is not performed because sufficient recognition accuracy cannot be obtained. However, since this process is merely a retry process for complementing the recognition result of the character recognition request area of the form format definition file 110, any area of the entire area in the form is invalid. Reading is not a problem.

In this case, the “item name” equivalent to the “item name” obtained from the character recognition request area registered in the form format definition file is replaced by the character recognition request area registered in the form format definition file 110. The "item name" is written in the form in advance, and there is little contact between the ruled lines and each character. Therefore, it is easier to recognize than "entry name". Word matching of “item name” using a word dictionary in which an arbitrary “item name” is registered (step 311)
Is performed, it can be considered that the "item name" can be sufficiently recognized even when the recognition is performed for all the areas in the form.

The word matching result detected in step 311 is referred to by the character recognition result analysis unit 105 as to whether it is equivalent to the item name acquired from the character recognition request area of the form format definition file 110 or by referring to the common knowledge database 112. A test is performed (step 312). If an item name equivalent to the item name acquired from the character recognition request area of the form format definition file 110 can be detected from the result of performing word matching on the recognition result of all areas in the form, the detected “item name” From the right area or the lower area, a character to be entered for "item name" is detected, and the recognition certainty of the character to be entered for "item name" is calculated (step 313).

Here, if the recognition certainty is high, it is registered as a candidate for complementing the recognition result. If the recognition confidence is low,
The character recognition result analysis unit 105 searches whether or not the character type of the character entered for “item name” for “the item name” is registered in the common knowledge database 112 (step 3).
14) If registered, the character recognizing unit 104 again inputs the character for the "item name" to the character type character dictionary 1
11, a character type dictionary corresponding to the character type is selected and recognized (step 315).

When the word matching result detected in step 311 has no recognition result character string equivalent to the item name detected in the character recognition request area of the form format definition file 110, the character recognition request of the form format definition file 110 is deleted. The recognition result of the input characters in the area and the recognition result of the input characters detected from the area other than the character recognition request area of the form format definition file 110 are integrated (step 316). Output (Step 317).

FIG. 4 is a diagram for explaining the form type determination in step 302 of FIG. A form 401 is an example of a form for performing form recognition. The form layout definition file 109 in FIG. 1 defines character strings indicating the characteristics of each form and a character recognition request area in which the character strings are described. 10
9B and 109C. In order to determine the form type, all form layout definition files 10
For 9A, 109B, and 109C, the character recognition request area defined is subjected to character recognition, and it is determined whether the recognition result is equal to the character string defined in the form layout definition files 109A, 109B, and 109C. To determine.

FIG. 5 is a diagram showing details of the item name word collation processing in step 305 in FIG. A collation process is performed on the character recognition result 502 of the item name “delivery amount” in the form 501 using the item name word collation dictionary 113 in which an arbitrary item name is registered, and the item name is acquired.

FIG. 6 shows an example of the form 601 and FIG.
3 shows details of the character type limitation processing of the character recognition request area registered in the form format definition file in steps 303 and 308 in FIG. 3 in the example of the form 601. When the form format definition file for the form 601 is 110 in FIG. 7, the upper left coordinates registered in the form format definition file 110 are (X1, Y1),
Character recognition 701 is performed on the area whose lower right coordinates are (X2, Y2) using the all-character-type dictionary 1114 in the character-type character dictionary 111. In the example of character recognition in FIG. 7, the “item name” is recognized as “full amount of the caretaker”, and the character entered for “item name” is recognized as “Y10.008”.

At this time, the recognition certainty of the entered character is calculated. In this case, since the recognition certainty is low, it is necessary to re-perform the recognition of the input characters for the “item name” with the character type limited. A character type to be used as preprocessing for re-executing a character to be entered for "item name" with character type limitation is determined. For this purpose, a word matching process is performed on the recognized item name “full payer” using the item name word matching dictionary 113 as shown in FIG. 5 to obtain the item name “delivery amount”.
Item name “delivery amount” obtained by item name word matching process,
Searched from the common knowledge database 112, "Delivery amount"
The character type "money character type" of the character to be entered is determined.

When the character type is determined, the character type character dictionary 11
1, the money amount character type dictionary 1113 corresponding to “money amount character type” is selected, and using this money amount character type dictionary 1113, the upper left coordinate (X
(3, Y3), character recognition 702 is performed on the area of the lower right coordinates (X2, Y2) to detect a character entered for “item name”, and a recognition result “$ 10,000” is output. You.

FIG. 8 shows an example of the form 801 and FIG.
3 shows details of the character recognition processing of all areas in the form in step 310 in FIG. 3 in the example of the form 801. In order to perform character recognition of the entire area of the form 801 in the form, first, line extraction 901 is performed, and character extraction 9 is performed for each line extraction result 904.
02 is performed. Character identification 90 for character extraction result 905
3 to obtain a character identification result 906.

FIG. 10 shows an example of a form 1001, and FIG.
Is a character string equivalent to the "item name" obtained from the character recognition request area of the form format definition file from the result of word matching in step 311 in FIG. 3 and the result of character recognition in the entire area in the form in the example of the form 1001 in FIG. This shows the details of the process of detecting the area and the area and acquiring the certainty factor of the character entered for “item name”.

The form format definition file 110 is a definition file corresponding to the form 1001. The all-area character recognition 1102 in the form is a result of performing all-area character recognition in the form 1101 in step 310 of FIG. As mentioned earlier, originally, character recognition processing for all areas in a form takes a long processing time, and line extraction is incorrect due to contact between frames and lines described in the form. This processing is not performed because sufficient recognition accuracy cannot be obtained, but this processing is merely a retry processing for complementing the recognition result of the character recognition request area of the form format definition file 802. It does not matter if the area is unread.

The character recognition request area registered in the form format definition file 110 is the form 1001 in FIG.
Area 1003. First, the recognition result 110 of the character recognition request area of the form format definition file 110
1 is subjected to an item name word matching process, and an item name matching result 502 is obtained. In order to obtain an item name having a meaning equivalent to the item name “address” of the item name collation result 502, a search is made as to whether the item name registered in the common knowledge database 112 has “address”, and the item name “ If there is an address, the item names “address”, “place”, and “ADDRESS” that are equivalent to “address” are detected.

Next, the item names “address”, “place”, “ADDRESS” acquired from the common knowledge database 112
Is searched from the item name / word matching result 1103 obtained by performing the item name / word matching process on the entire area character recognition result 1102 in the form. In the example of FIG. 11, "address" and "place" are detected. However, since the detection area for the item name "address" is the same area as the character recognition request area registered in the form format definition file 110. Do not adopt. The detection area of the item name "Oko" is the form format definition file 1
Since it is different from the character recognition request area of No. 10, the degree of certainty of the recognition result of the entered character for the “item name” in the right area or the lower area of the “item name” detection area is obtained from the entire area character recognition result 1102 in the form. .

FIG. 12 shows details of the recognition result integration processing in step 316 of FIG. In an example of a form 1201 in which a plurality of entry characters having the same meaning are described in different areas, from the entry character recognition result 1202 of the character recognition request area registered in the form format definition file 110 and the form format definition file character recognition request area "item name"
When the result 1203 of recognizing the input character corresponding to "item name" by detecting "the item name" and "equivalent item name" from an arbitrary area in the form and recognizing the characters to be entered is recognized for each character. The certainty factors are compared, and the one with the higher recognition certainty factor is integrated as a result to obtain an entry character integrated result 1204.

As described above, in the prior art, if the confidence of the entered character for the "item name" is low in step 304 in FIG. 3, the recognition is impossible or misread. By using the method, even if the degree of certainty of the input character for “item name” is low (step 304 in FIG. 3), the “item name” is recognized, and the equivalent input character is automatically extracted from another area based on the result. By detecting and integrating the recognition results, higher recognition accuracy can be obtained. Also, by recognizing the "item name" and selecting the character type of the entry character for the "item name" from the common knowledge database 112, the character type of the entry character for the "item name" is registered for each form format definition file 110. Therefore, it is possible to reduce the load on the operator who creates the form format definition file.

[0036]

According to the present invention, a form recognition method capable of improving recognition accuracy can be obtained. Further, according to the present invention, it is possible to obtain a form recognition method capable of reducing the work of an operator.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing an embodiment of a form recognition apparatus to which a form recognition method of the present invention is applied.

FIG. 2 is a diagram illustrating an example of a form, an example of a form format definition file corresponding thereto, an example of a common knowledge database, and an example of a character type character dictionary.

FIG. 3 is a flowchart showing a flow of processing of the entire form recognition apparatus of FIG. 1;

FIG. 4 is a diagram for explaining form type determination in step 302 in FIG. 3;

5 is a diagram showing details of an item name word matching process in step 305 of FIG. 3;

FIG. 6 is a diagram illustrating an example of a form.

7 is a diagram showing details of a character type restriction process of a character recognition request area registered in a form format definition file in steps 303 and 308 in FIG. 3 in the example of the form 601 in FIG.

FIG. 8 is a diagram illustrating an example of a form.

9 shows an example of the form shown in FIG. 8;
FIG. 11 is a diagram showing details of a character recognition process for all areas in a form of 0.

FIG. 10 is a diagram illustrating an example of a form.

11 is equivalent to “item name” obtained from the character recognition request area of the form format definition file from the word collation result of step 311 in FIG. 3 and the character recognition result of all areas in the form in the example of the form in FIG. FIG. 14 is a diagram illustrating details of processing for detecting a character string and an area and acquiring the certainty factor of a character entered for “item name”.

FIG. 12 is a diagram illustrating details of a recognition result integration process in step 316 of FIG. 3;

[Explanation of symbols]

101: form recognition unit, 102: image scanner, 10
3: Form identification unit, 104: character recognition unit, 105: character recognition result analysis unit, 106: recognition result integration unit, 107: recognition result output unit, 108: host application, 109:
Form layout definition file, 110: Form format definition file, 111: Character dictionary, 112: Common knowledge database, 113: Item name word collation dictionary, 2
01,401,501,601,801,1001,1
201: Report.

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Eisuke Asano 1 Ikegami, Haruoka-cho, Owariasahi-shi, Aichi F-term in the Information Equipment Division, Hitachi, Ltd. DA34 EA17 EA19 EA27

Claims (2)

[Claims] 1. A form recognition means for limiting a form type to an input form image, performing character recognition on the form image with the limited form type, and outputting a recognition result to a host device. When performing character recognition on the recognition request area in the form image,
A form recognition method, comprising detecting a character string equivalent to a character string described in the recognition request area from an area other than the recognition request area, integrating recognition results, and outputting the result to the host device. 2. The form recognition method according to claim 1, wherein, in the limited form type, when performing character recognition on a recognition request area in the form image, a character type used in character recognition is limited. A form recognition method characterized by performing recognition.