JP2001060249A - Character string collation system for document identification - Google Patents

Abstract

PROBLEM TO BE SOLVED: To increase the character string collation speed of a document identifying process for identifying the kind of a document which does not have a dedicated identification code for document identification by collating the character string entered into the document. SOLUTION: This system is equipped with a blank decision means 32 which decides whether or not a pattern of a character, etc., is included in a cut partial image and a character size decision means 33 which detects a rectangle circumscribed with the pattern in the cut partial image and compares the size of the rectangle with information on the sizes of rectangles circumscribed with respective characters M stored in a dictionary to decide whether the sizes of both the rectangles match each other. When the cut partial image is blank or the size of the pattern included in the cut partial image is different from the size of the rectangle circumscribed with the character M matching the pattern, collation against the character string including the character M is immediately quit and collation against a next character string is started.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a form identification character string collating method, and more particularly to a form identification character string collating method for identifying the type of an existing form on which an identification code for an optical character reader is not printed.

[0002]

2. Description of the Related Art Conventionally, when a plurality of types of forms are processed in an optical character reading device, a predetermined position is fixed in each form in advance.
Write the ID (identification) number by printing or handwriting.
In general, by reading this ID number, a character reading process is performed with reference to the form format information registered corresponding to each form type. In this case,
Existing forms without ID numbers cannot be processed.

[0003] Japanese Patent Application No. 10-081907 describes a feature extracted from a character string image such as a form name described at a specific position on a read form and a character string registered in a dictionary in advance. The form type is identified by collating with a feature amount that should be.

As a reference technique, Japanese Patent Application Laid-Open No. 5-274 is disclosed.
No. 471 discloses a "title area extraction processing method for image documents" in which a title area is determined and extracted from an input image and a title of image data is assigned without relying on a mark sheet or keyboard input. JP-A-9-1
Japanese Patent No. 34406 discloses a method of setting a rectangular area circumscribing continuous characters from a document image captured as image data, and evaluating attributes such as character height, emphasis, and ruled lines to extract title candidates. Title Extraction Apparatus and Method ". Also, Japanese Patent Application Laid-Open No. Hei 10-11531 discloses that
When reading a blank form and creating a form format,
Disclosed is a "form reader" in which a character string having a characteristic in character size, style, or the like is selected as a title candidate from the identification result of a preprinted character string printed in advance.

[0005]

SUMMARY OF THE INVENTION In the prior art,
There is a problem that the time required for identification increases as the number of forms and characters registered in the dictionary increases. The reason is that as the number of registered form types increases,
Further, as the number of characters in the character string increases, the number of times that each character in the character string of each form type is matched with the character image cut out from the input image increases.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a form identification character string collating method capable of collating a character string described on a form image at high speed and identifying a form type of an input form.

[0007]

In order to solve the above-mentioned problems, a character string collating method for form identification according to the present invention employs the following characteristic configuration.

(1) In a form identification character string collation method for collating a character string described for identifying the form at a predetermined position of the form, characters are included in a partial image obtained by cutting out a predetermined position of the form. Whether or not the character is included is determined by the blank determining means, a circumscribed rectangle of the character in the partial image is detected, and the circumscribed rectangle stored in the dictionary storage means is compared by the character size determining means to determine a match. Form identification character string collation method.

(2) The form identification character string collation method according to (1), wherein the determination is individually performed for each character of the character string.

(3) The form identification character string collation method according to the above (1) or (2), wherein if the result of the determination is a mismatch, the collation work is immediately stopped and the process proceeds to the next collation.

(4) The form identification character string collation method according to (1), wherein the blank determining means binarizes the cut-out partial image based on a preset threshold value.

(5) an image input device for capturing an image of a form, an image storage unit for storing the captured image, and a storage device having a dictionary storage unit for storing character position information and character characteristic information; Character image cutout means for cutting out a character image from a captured image, blank space determination means for performing blank space determination,
A form identification character string collating method comprising: a character size determining means for determining a character size; and a data processing device including a character collating means for collating characters based on information in the storage device.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of a form identification character string collating method according to the present invention will be described below in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram showing the configuration of a preferred embodiment of a form identification character string collating method according to the present invention. This form identification character string collation method uses an image input device.
1. It comprises a storage device 2 and a data processing device 3.
The storage device 2 includes an image storage unit 21 and a dictionary storage unit 22. The dictionary storage unit 22 stores character position information 22a and character characteristic information 22b in association with each other. In addition, the data processing device 3 includes a character image cutout unit 31, a blank determination unit 32,
It includes a character size determination unit 33 and a character collation unit 34.

The image input device 1 is a scanner or the like. The storage device 2 stores information. The data processing device 3 operates under program control. More specifically,
The image input device 1 reads an image of a form and takes it into the data processing device 3 as a multi-tone two-dimensional image. The image storage unit 21 of the storage device 2 stores an image of a form captured by the image input device 1. The dictionary storage unit 22 stores data for collating a character string. Character position information 22a, which stores the position of each character constituting a character string on a form and the size of a rectangle circumscribing the character for each form type, and character characteristic information 22b are stored in association with each other. In the character feature information 22b, a feature amount of a character string serving as a matching power is stored for each character. As the feature amount, each character image is divided into a grid-like partial region, and the average of the pixel values in each partial region arranged for the number of divisions is used as the feature amount.
Other features may be used.

The character image extracting means 3 of the data processing device 3
1 cuts out a partial image at a position where a character is to be written from the form image stored in the image storage unit 21 using the above-described character position information 22a. The blank determining unit 32 determines whether a figure such as a character exists in the cut-out partial image. The character size determination unit 33 determines whether the size of the graphic included in the cut-out partial image matches the size of the character to be cut out. The character matching unit 34 extracts a feature amount from the cut-out partial image and matches the feature amount with the feature amount stored in the above-described character feature information 22b.

Next, the operation of the form identification character string collation system of FIG. 1 will be described with reference to the flowcharts of FIGS. FIG. 2 is a flowchart showing the overall operation of the form identification character string collation method of FIG. FIG. 3 is a flowchart of the blank determining unit 32 in FIG. FIG.
3 is a flowchart of the character size determination means 33 in FIG.

First, an image of a form is read as a two-dimensional image of multi-gradation by the image input device 1, and the image is stored in the image storage unit 2.
1 (step S1 in FIG. 2). Next, the data processing device 3 selects one of the form types stored in the dictionary storage unit 22 (Step S2 in FIG. 2). Further, the data processing device 3 selects one character from the beginning of the character string corresponding to the selected form type (step S3 in FIG. 2). Next, based on the character position information 22a, the character image cutout means 31 extracts the partial image MJI as shown in FIG. 5 at the position where the character selected in step S3 should be described from the form image stored in the image storage unit 21. (Figure 2)
Step S4). Next, the blank determining means 32 determines whether or not a graphic such as a character is included in the image MJI cut out in step S4 (step S5 in FIG. 2).

Next, the operation of the blank determining means 32 will be described with reference to the flowchart of FIG. First, a threshold value (threshold) T for binarizing the image MJI is obtained (step S51 in FIG. 3). This binarization threshold T is
1980, IEICE Transactions Vol.J63-D, No.
4, may be obtained by the discriminant analysis method proposed by Otsu in pp. 349-356. Otsu's method is a technique of selecting a threshold value in an image including two categories of characters and a background so as to minimize the variance within each category and maximize the variance between categories. Next, as shown in FIG. 6, for example, an average value m1 of the pixel values less than the binarization threshold value T obtained in step S51 and an average value m2 of the pixel values equal to or more than the binarization threshold value T are obtained (FIG. Third step S52). After that, the evaluation value Es = | m1-m2 | is calculated (step 53 in FIG. 3). Here, the magnitude of the evaluation value Es is based on the partial image MJI
Represents the degree of division when the pixel values in are divided into two by the binarization threshold value T. A large evaluation value Es indicates that the image MJI has a graphic part such as a character and a background part.

The blank determining means 32 then proceeds to step S53
It is determined whether or not the evaluation value Es calculated in is smaller than or larger than a preset constant TH (step S54 in FIG. 3).
It is necessary to determine an appropriate value for the constant TH according to the number of gradations of the image to be handled. For example, when the number of gradations of the image is 256, TH = 20. Evaluation value E in step S54
If it is determined that s is smaller than the constant TH, the partial image M
It is assumed that no graphic such as a character exists in JI, and the blank determination flag FGSP is set to “1” (step S55 in FIG. 3). On the other hand, when it is determined in step S54 that the evaluation value Es is equal to or larger than the constant TH, it is determined that a graphic such as a character exists in the image MJI, and the blank determination flag FGSP is set to “0” (step S56 in FIG. 3). ). Finally, the blank determination unit 32 passes the value of the blank determination flag FGSP to the data processing device 3.

Next, returning to FIG. 2, the description of the embodiment of the form identification character string collation method of FIG. 1 will be continued. The data processing device 3 determines whether the blank determination flag FGSP is “1” or “0” (Step S6 in FIG. 2). If the blank determination flag FGSP is determined to be "1" in step S6, the process proceeds to step S12. On the other hand, when the blank determination flag FGSP is determined to be "0" in step S6, the process proceeds to step S7. Then, the character size determination means 33 determines the size of the rectangle RCD circumscribing the character selected in step S3,
It is determined whether or not the size of the rectangle RCI circumscribing the figure included in the image MJI matches (step S7 in FIG. 2).

Next, the operation of the character size determining means 33 will be described with reference to the flowchart of FIG. First, the image MJ is calculated using the binarization threshold value T obtained in step S51 of FIG.
I is binarized to obtain a binary image value MJI2 (step S in FIG. 4).
71). The binarization process in this step S71 is performed based on the image MJI
If the pixelization of the upper target pixel is equal to or larger than the binarization threshold T, “1” is obtained by writing “0” to the position of the target pixel of the image MJI2 if smaller. However, here the multi-valued image MJ
In I, it is assumed that a pixel having a high pixel value corresponds to a graphic portion such as a character, and a pixel having a low pixel value corresponds to a background portion.

The character size judging means 33 then proceeds to FIG.
As shown in (A), the size of the rectangle RCI circumscribing a figure (character) existing in the binary image MJI2, that is, the width WI and the height HI are detected (step S72 in FIG. 4). This step S72
The detection of the rectangle circumscribing the figure in is performed as follows. First, the image is scanned in the X (horizontal) axis direction from the upper end of the binary image MJI2, and the Y (vertical) axis coordinate y1 at which the pixel value “1” first appears is detected. Next, X from the lower end of the binary image MJI2
The image is scanned in the axial direction, and the pixel value “1” appears first in Y
The axis coordinate y2 is detected. Next, the image is scanned in the Y-axis direction from the left end of the binary image MJI2, and the X-axis coordinate x1 where the pixel value “1” first appears is detected. Further, the image is scanned in the Y-axis direction from the right end of the binary image MJI2, and the X-axis coordinate x2 where the pixel value “1” first appears is detected. The width WI = x2-x of the rectangle circumscribing the figure in the binary image MJI2 in step S72 from the obtained x1, x2, y1, and y2.
1 + 1 and HI = y2-y1 + 1 are obtained.

Next, the character size determining means 33 determines the size of the rectangle RCD circumscribing the character selected in step S3, that is, the width WD and the height HD, and the rectangle circumscribing the figure in the binary image MJI2. Using the size of the RCI, that is, the width WI and the height HI, it is evaluated whether the following expression is satisfied (step S73 in FIG. 4). | WD-WI | <WD / 4 and | HD-HI | <HD / 4 (1) If the above expression is not satisfied in step S73, step S
Size and image MJI of rectangle RCD circumscribing the character selected in step 3
It is determined that the sizes of the rectangles circumscribing the figure included in the box do not match, and the character size determination flag FGSZ is set to “1”.
Is substituted (step S74 in FIG. 4). If the above expression is satisfied in step S73, it is determined that the size of the rectangle RCD circumscribing the character selected in step S3 and the size of the figure included in the image MJI match, and the character size flag is set. "0" is substituted for FGSZ (step S75 in FIG. 4).
Finally, the character size determination unit 33 passes the value of the character size determination flag FGSZ to the data processing device 3.

The data processing device 3 determines whether the value of the character size determination flag FGSZ is "1" or "0" (step S8 in FIG. 2). In this step S8, the character size determination flag
If the value of FGSZ is determined to be “1”, step S12
Proceed to processing. On the other hand, if the value of the character size determination flag FGSZ is determined to be “0” in step S8,
The process proceeds to S91.

The character collating means 34 cuts out the characters in the partial image MJI cut out in step S4 according to the circumscribed rectangle RCI, and extracts a characteristic amount from the cut out characters (step S91 in FIG. 2). As the feature amount, the same feature amount as when the character feature information 22b of the dictionary storage unit 22 is created is used. Next, the character matching unit 34 matches the extracted character feature amount with the character matching feature amount stored in the character feature information 22b of the dictionary storage unit 22 according to the following equation, and obtains the simple similarity S
(I) is obtained (step S92 in FIG. 2). Si = (a · b) / (| a | × | b |) where (a · b) is the inner product of vector a and vector b, | a |
Denotes the norm of the vector a.

The data processing device 3 determines whether or not all characters in the character string corresponding to the form type selected in step S2 have been collated (step 1 in FIG. 2).
0). As a result, if the collation processing of all the characters has not been completed, the next character to be collated is selected (step S
3), the processes of steps S4 to S10 are repeatedly executed. If the collation processing of all characters has been completed, the average value of S (i) is obtained, and the average value is determined as the similarity of the character string corresponding to the form type selected in step S2 (step S11 in FIG. 2).

The data processing device 3 executes step S6 or S8
If the value of the blank determination flag FGSP or the character size determination flag FGSZ is determined to be “1” in step
The similarity of the character string corresponding to the form type selected in step 2 is "0"
(Step S12 in FIG. 2). Next, the data processing device 3 determines whether all the form types have been selected in step S2 (step S13). As a result, if there is an unselected form type (No), the next form type to be collated is selected in step S2, and the processing in steps S3 to S13 is executed. On the other hand, all report types were selected (Yes)
In this case, the form ID for identifying the form type corresponding to the character string having the highest similarity is output as the input form type (step S14).

Next, the operation of the preferred embodiment of the present invention will be described based on a specific example. It is assumed that the form shown in FIG. 8 has been input. First, the image input device 1 captures a form image as a two-dimensional multivalued image and stores it in the image storage unit 21 of the storage device 2 (step S1). Next, the data processing device 3
Selects one of the different types of forms as shown in FIG. 9 stored in the dictionary storage unit 22 (step S2). For example, it is assumed that “tax report” is registered as a character string corresponding to the selected report type. Next, the data processing device 3
Selects one character in the character string "tax form" corresponding to the selected form type, here, "tax" which is the first character (step S3).

Therefore, the character image extracting means 31 uses the character position information 22a of "tax" stored in the dictionary storage unit to extract the input form image stored in the image storage unit 21 from the input form image.
A partial image MJI as shown in FIG. 5 having the position and size where "tax" should be described is cut out (step S4). In order to determine whether or not the cut-out partial image MJI includes a graphic such as a character, the process proceeds to the process of the blank determining means 32 (step S).
5). The blank determining means 32 first obtains a binarization threshold T for binarizing the partial image MJI by a discriminant analysis method or the like (step S51). Next, the average value m1 of the pixel values less than the binarization threshold value T and the average value m2 of the pixel values higher than the threshold value T are obtained (step S52), and the blank judgment evaluation value Es is calculated (step S53). Here, assuming that the blank judgment evaluation value Es is equal to or larger than the constant TH, that is, if a figure is included in the partial image MJI, the blank judgment unit 32
"0" is substituted for FGSP (step S56).

Next, the data processing device 3 determines whether the value of the blank determination flag FGSP is "1" or "0" (step S).
6). Since the value of the blank determination flag FGSP is “0”, the process proceeds to step S7. Therefore, the character size determination unit 33 determines whether the size of the rectangle RCD circumscribing the character "tax" selected in step S3 and the size of the rectangle RCI circumscribing the figure included in the partial image MJI match. It is determined whether or not it is (step S7). The character size judging unit 33 first calculates the binarized threshold value obtained by the blank judging unit 32.
The partial image MJI is binarized by T to obtain a binary image MJI2 (step S71). Next, the width WI and the height HI of the rectangle RCI circumscribing the figure included in the binary image MJI2 are detected (step S72). Next, the size of the circumscribed rectangle RCI and the dictionary storage unit 2
Whether or not the size of the circumscribed rectangle RCD of “tax” stored in the character position information 22a in 2 is the same is determined by the above formula (step S73). When this condition is satisfied (satisfied), it is assumed that the sizes of both circumscribed rectangles match. The character size determination unit 33 substitutes “0” for the character size determination flag FGSZ (step S75).

Next, the data processing device 3 determines whether the value of the character size determination flag FGSZ is "1" or "0" (step S8). Since the value of the character size determination flag FGSZ is “0”, the process proceeds to step S91. The character matching unit 34 determines whether the partial image MJI
Are extracted in accordance with the circumscribed rectangle RCI obtained in step S72, and a feature amount is extracted from the extracted characters (step S72).
S91). Next, the character matching unit 34 compares the extracted character feature amount with the matching feature amount of the character stored in the character feature information 22b of the dictionary storage unit 22 according to the above-described formula, and compares the extracted character feature amount with the character “tax”. The simple similarity S (i) is obtained (step S92).

Therefore, the data processing device 3 executes step S
It is determined whether all the characters in the character string “tax form” corresponding to the form type selected in step 2 have been collated (step S1).
0). Here, since "book" and "vote" have not been referred to yet, the process returns to step S3 and selects the character "book".
Thereafter, the same processing as described above for "tax" is performed on this "book" up to step S5. Here, it is assumed that the blank determination unit 32 determines that a graphic is included in the partial image MJI at the position where “book” should be described. Then, the data processing device 3 shifts the processing to the character size determining means 33 (step S7). Character size determination means 33
It is assumed that it is determined that the size of the circumscribed rectangle RCI does not match the size of the circumscribed rectangle RCD of “book” stored in the character position information 22 a in the dictionary storage unit 22. Then, as a result of determining the value of the character size determination flag FGSZ in step S8, the data processing device 3 shifts the processing to step S12 because the value of FGSZ is “1”, and the data processing device 3 determines the form type selected in step S2. The similarity of the corresponding character string “tax report” is set to “0”.

Next, the data processing device 3 executes step S1.
In step 3, it is determined whether all the form types have been selected. Here, it is assumed that an unselected form type exists in the dictionary storage unit 22 in step S2, and the process returns to step S2 again to select the next form type. The data processing device 3 selects the first character “pay” of the character string “payment note” corresponding to the form type selected in step S2 (step S3). Partial image MJI at the position where this character should be described
Is cut out from the input form image (step S4). Therefore,
The blank determining unit 32 determines whether a graphic is included in the partial image MJI (step S5). Here, it is assumed that since the blank judgment evaluation value Es calculated in step S53 is less than the constant TH, it is determined that no graphic is included in the partial image MJI. Then, the data processing device 3 determines the value of the blank determination flag FGSP in step S6,
Since the value of SP is “1”, the process proceeds to step S12, and the similarity of the character string “payment note” corresponding to the form type selected in step S2 is set to “0”.

Next, the data processing device 3 executes step S1.
In step 3, it is determined whether all the form types have been selected. Here, it is assumed that an unselected form type exists in the dictionary storage unit 22 in step S2, and the process returns to step S2 again to select the next form type. The data processing device 3 executes the step S2 for each of the characters "tsu", "knowledge" and "letter" constituting the character string "notification" corresponding to the form type selected in step S2.
Steps 3 to S10 are executed. Here, it is assumed that the processing in step S12 has not been performed from the branch in step S6 and the branch in S8 for all of the characters "", "", and "". Then, in step S92, the similarity S of each character is determined for each of the characters "", "", and "", and the average value of these three similarities is calculated as the similarity of the character string "notification letter". I do. Thereafter, the data processing device 3 sets the dictionary storage unit 22a
Step that all form types stored in
Until it is confirmed in S13, steps S2 to S13
Is repeated.

When it is confirmed in step S13 that all the form types have been selected, the form type corresponding to the character string having the highest similarity S among all the form types stored in the dictionary storage unit 22a is determined. The form ID shown is output as the form type of the input form, and the process is terminated (step S14). here,
If the similarity of the string "notice letter" was the highest,
The data processing device 3 outputs a form ID of a form type corresponding to the character string “notification letter”.

[0037]

As is clear from the above description, according to the present invention, if no character is described at the position where the character constituting the character string to be collated on the input form should be described, If the size of the graphic described in the position is different from the size of the character to be collated, the collation processing of the character string is immediately stopped, and the processing proceeds to the collation processing of the next character string. For this reason, the character collation process (steps S3 to S3) executed during collation for all character strings
92), the time required for form identification can be shortened. Further, it is possible to prevent the form type corresponding to the above-described character string from being a candidate for the form type of the input form.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a preferred embodiment of a form identification character string collating method according to the present invention.

FIG. 2 is a flowchart for explaining the operation of the form identification character string collation method of FIG. 1;

FIG. 3 is a flowchart showing a blank determination procedure of a blank determination unit in FIG. 1;

FIG. 4 is a flowchart showing a character size determination procedure of a character size determination unit in FIG. 1;

FIG. 5 is a diagram illustrating an example of a partial image MJI.

FIG. 6 is a density histogram used in a blank determination procedure of a blank determination unit in FIG. 1;

FIG. 7 shows (A) a figure circumscribed rectangle RCI and (B) a character circumscribed rectangle RCD used in the character size determination procedure of FIG.

FIG. 8 shows an example of a form.

FIG. 9 shows examples of different types of forms stored in the dictionary storage unit in FIG. 1;

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 image input device 2 storage device 3 data processing device 21 image storage unit 22 dictionary storage unit 31 character image cutout unit 32 space determination unit 33 character size determination unit 34 character collation unit

Claims (5)

[Claims] 1. A form identification character string collating method for collating a character string described for identifying the form at a predetermined position of the form, wherein a character is included in a partial image obtained by cutting out a predetermined position of the form. The circumscribed rectangle of the character in the partial image is detected, and the circumscribed rectangle stored in the dictionary storage unit is compared with the circumscribed rectangle stored in the dictionary storage unit to judge whether or not there is a match. A character string collating method for form identification characterized by the following. 2. The form identification character string collating method according to claim 1, wherein said determination is made individually for each character of said character string. 3. The form identification character string collating method according to claim 1, wherein if the result of the determination is a mismatch, the collating operation is immediately stopped and the next collation is performed. 4. The form identification character string according to claim 1, wherein the blank determining means performs the binarization of the cut-out partial image based on a preset threshold value. Matching method. 5. An image input device for capturing an image of a form, a storage device having an image storage unit for storing the captured image, and a dictionary storage unit for storing character position information and character characteristic information. Character image cutout means for cutting out a character image from an embedded image, blank space determination means for performing space determination, character size determination means for determining character size, and character collation means for collating characters based on information in the storage device And a data processing device including a character string collating method for form identification.