JP2007079966A - Character recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform character recognition about a portion having small distortion of a business form when acquiring an image in a state that the business form floats from a mat. <P>SOLUTION: This character recognition device acquiring the image of the business form, and performing the character recognition according to the distortion of an outer shape of a business form portion on the image has: a distortion characteristic extraction part extracting a distortion characteristic of the outer shape of the business form portion; a business form distortion characteristic DB wherein a distortion pattern is classified according to largeness of the distortion in each the portion of the business form; a distortion identification part selecting the corresponding distortion pattern on the basis of the distortion characteristic of the extracted business form portion; and a character recognition part performing the character recognition according to the selected distortion pattern. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a character recognition apparatus that performs character recognition by determining distortion of a form image using a non-contact scanner device that sets a form on a mat and reads an image from the top with a sensor unit.

Currently, non-contact scanners for acquiring images of various papers such as forms and identification documents brought in by customers are used in the counter business of financial sales offices. The non-contact type scanner is a type of scanner that acquires an image of a form by a camera installed above a mat portion on which the form is placed. In the counter business of a financial sales office, the non-contact type scanner is connected to the counter terminal of the financial sales office counter, and image acquisition control is performed. The non-contact type scanner has features such as a simpler operation for image acquisition and a shorter time required than a flat bed type scanner or a sheet feed type scanner. (For example, Non-Patent Document 1)
The form image acquired at the window is subjected to predetermined character recognition processing by an optical character recognition device, and the contents of entry are converted into data and used for financial transactions.
The prior art of the present invention will be described below.

<Conventional system configuration>
An overview of a conventional system is shown in FIG.
The conventional system includes a non-contact scanner device 101, an optical character recognition device 102, and a WS 103. The WS 103 includes input means such as a keyboard and display means such as a CRT.
In FIG. 6, the non-contact scanner device 101 is viewed from above. The non-contact scanner device 101 and the optical character recognition device 102 will be described later.
The WS 103 is a terminal for controlling the entire system, and a general PC or the like having a display unit such as a memory and a CRT (not shown) and an operation unit such as a keyboard mouse is used.
Here, the optical character recognition apparatus 102 is described as being realized by hardware, but may be software provided in the WS 103.

<Configuration of non-contact scanner device>
As shown in FIG. 7, the non-contact scanner device 101 includes a control unit 101c, a mat unit 101a, and an image acquisition unit 101b.
The control unit 101 c includes a control unit for the entire non-contact scanner device 101 and a communication unit with the optical character recognition device 102.
The mat unit 101a is a table for installing a form for acquiring an image. The mat portion 101a is represented by shading in FIG. 6, but a material of a color such as black that can be easily distinguished from the form on the image is used. As a result, the acquired form image has a mat portion as a background, so that the form appears to rise against the black background color. For this reason, it is possible to easily detect the sides of the form.
The image acquisition unit 101b is a camera using a CCD, a C-MOS, or the like that has a function of photographing a form on the mat unit 101a and converting it into image data.

<Configuration of optical character recognition device>
The optical character recognition device identifies the type of form from the acquired form image, further identifies the area of the form to be recognized, and performs character recognition.
As shown in FIG. 8, the configuration of the optical character recognition device 102 includes a control unit 102a, an image memory 102b, a form angle detection unit 102c, a format creation unit 102d, a format reading unit 102e, a character area correction unit 102f, and a character recognition unit. 102g. In addition, communication means (not shown) with the non-contact scanner device 101 and the WS 103 is included.

The control unit 102 a of the optical character recognition device 102 controls the entire optical character recognition device 102.
The image memory 102b has means for holding the form image data. Each processing unit can refer to the form image by appropriately referring to the image memory 102b.
The form angle detection unit 102c has means for detecting a form angle from the image. The corner of the form is detected by, for example, scanning an image and detecting the boundary between the background and the form part, that is, the form side based on the difference in pixel density. By tracking the obtained pixels as the sides of the form, it is possible to determine whether the change of the X or Y coordinate is reversed or the place where the change starts is a corner. Alternatively, a plurality of points can be arbitrarily selected from each form side, and a known technique such as Hough transform can be used to approximate each form side to a straight line or a curve, and the intersection point can be selected as a corner.
The format creation unit 102d has means for creating a format necessary for character recognition such as character region coordinates. The format creation unit 102d will be described later.
The format reading unit 102e provides a function for reading the contents of the format to other processing units.
The character area correction unit 102f has means for correcting the character area coordinates from the format. The character area correction unit 102f will be described later.
The character recognition unit 102g provides a function of recognizing characters by pattern matching with a character recognition dictionary (not shown).

<Format creation section>
As shown in FIG. 9, the format creation unit 102d includes a control unit 102d1, a format storage unit 102d2, and a character area coordinate setting unit 102d3. Moreover, it has a communication processing unit (not shown) with other processing units.

The control unit 102d1 of the format creation unit controls the entire format creation unit 102d.
The format storage unit 102d2 provides a function of storing a format necessary for character recognition such as character region coordinates in a format storage unit (not shown). Hereinafter, the format is acquired from the format storage unit.
The format stores relative coordinates from the corner (for example, upper left), which is the basis of the form, for each character area, along with the identification information of each form. Based on this information, the character recognition unit identifies the character area on the form image and recognizes the entered character.
The character area coordinate setting unit 102d3 receives the input operation of the operator who creates the format, and registers necessary information such as the type of dictionary used for character recognition in addition to the above coordinates as the format. Here, it is assumed that the operator inputs data to the format creation unit via the WS 103.

<Character area correction unit>
As shown in FIG. 10, the configuration of the character region correction unit 102f includes a control unit 102f1 and a character region position correction unit 102f2. Moreover, it has a communication processing unit (not shown) with other processing units.
The control unit 102f1 of the character area correction unit 102f controls the entire character area correction unit 102f.

  The character region position correction unit 102f2 corrects the character region from the coordinates of the upper left corner. Note that although the upper left corner is used here, the reference angle of the relative coordinate is the upper left corner. For example, all other corners such as the upper right corner can be used as a reference.

<Operation>
(A) Format Registration FIG. 11A is a flowchart at the time of registration in an optical character recognition system using a non-contact scanner. First, the image acquisition unit 101b reads a registered form placed on the mat unit 101a, and captures the acquired image into the image memory 102b (S101). Subsequently, the form angle detection unit 102c refers to the image memory 102b to detect four form angles on the image (S102). Subsequently, the character area coordinate setting unit 102d3 refers to the image memory 102b to display an image on the CRT of the WS 103, and the operator sets the character area (S103). Subsequently, the format storage unit 102d2 stores the form corner coordinates and the character area coordinates in the format together with the identification information of each form such as a sheet ID specified separately (S104).

(B) Reading time FIG. 11B is a flowchart at the time of reading in the optical character recognition system using the non-contact scanner. First, the image acquisition unit 101b reads a read form placed on the mat unit 101a, and captures the acquired image into the image memory 102b (S201). Subsequently, the form angle detecting unit 102c refers to the image memory 102b to detect four form angles on the image (S202). Subsequently, the format reading unit 102e reads the format corresponding to the type of the form (form corner coordinates and character area coordinates of the registered image) (S203). Subsequently, the character area correction unit 102f2 of the character area correction unit 102f corrects the character area from the form angle coordinates / character area coordinates of the registered image and the form angle coordinates of the read image by the upper left corner coordinates of the form angle coordinates. (S204). Subsequently, the character recognition unit 102g recognizes the character (S205), and displays the character recognition result on the CRT of the WS 103 (S206). An image obtained by cutting out the character area is displayed as a confirmation image together with the character recognition result, and is often used for confirmation of the character recognition result.

The type of the form is specified by comparing the identification information extracted from the form image with the identification information registered in the format and selecting a matching format. As the identification information, a sheet ID, ruled line feature information, or the like is used.
Oki Technical Review January 2002 / No. 189 Vol. 69 No. 1 “Color Direct Scanner”

  As described above, the non-contact scanner has a structure in which an image is acquired by photographing a form placed on the mat portion with a camera from above, so when the image is acquired with the form raised from the mat, the form itself is Even if it is not distorted, the form is distorted on the image.

FIG. 12B expresses the form distortion when the image is acquired with the upper left corner of the form raised. In such a case, since the corner of the character area of the form is shifted so that it extends to the upper left, when character recognition is performed, the character area is erroneously moved to the upper left as shown in the figure. Will correct. Since all character area positions on the form are determined with reference to the corner of one form, the other character areas are also shifted uniformly. For this reason, the entire form has an influence such as misreading of the character recognition result, blanking, or dropping of digits due to an actual entry character being out of the character area. In such a case, the confirmation image of the character area is also cut out in a shifted manner, so that it is not only difficult to correct, but also difficult to notice digit loss. For example, the character area in the lower right part of FIG. 12B is completely deviated from the actual entry position because the reference corner is shifted. In such a case, it may be regarded as blank even though there is actually an entry. In such cases, financial transaction accidents may occur.
For this reason, when such distortion is detected, it is safe to reject the cause of the form to remove the cause of the distortion and perform the same operation again.

By the way, the lower right portion of FIG. 12B is actually hardly distorted, and the character region is shifted due to the distortion of the upper left portion. In such a case, it is efficient to perform character recognition for the portions other than the upper left. For example, if an explanatory note to the writer is printed in the upper left part and there is a character recognition area in other parts, the upper left part is a part that does not perform character recognition. If the same operation is requested again due to distortion, the work efficiency is poor.
Therefore, when a form image in such a state is acquired, it is efficient to perform character recognition on a part with less distortion of the form.

In order to solve such a problem, the character recognition device of the present invention has the following configuration.
<Configuration 1>
A character recognition device that acquires an image of a form and performs character recognition in accordance with the distortion of the outline of the form part on the image, a distortion feature extraction unit that extracts the distortion characteristic of the outline of the form part, and a distortion A form distortion feature DB that classifies patterns according to the size of distortion for each part of the form, a distortion identification unit that selects a corresponding distortion pattern based on the distortion characteristics of the extracted form part, and a selected distortion pattern A character recognition unit that performs character recognition accordingly is provided.
<Configuration 2>
The character recognition unit performs character recognition on a portion with a small distortion according to the selected distortion pattern.
<Configuration 3>
The character recognition unit selects a character region coordinate correction method according to a distortion pattern.
<Configuration 4>
An edge dividing unit that obtains an edge dividing point obtained by dividing an edge of the form part, and the distortion feature extracting unit connects the edge dividing points of the opposing sides according to a predetermined correspondence condition, and the line segments at the intersection of each line segment Are extracted as distortion features.
<Configuration 5>
A corner detection unit that detects a corner of the form part, wherein the side division unit obtains points obtained by dividing the lines connecting the detected corners, and associates the obtained points with the sides of the form part to obtain sides; Split.
<Configuration 6>
The side dividing unit divides the division equally into a predetermined number.
<Configuration 7>
The side dividing portion shortens the interval of the division in a portion close to a corner.
<Configuration 8>
The interval of the division is determined by mapping the position of the division point when equally divided by a trigonometric function.
<Configuration 9>
In the form distortion feature DB, a distortion pattern for each form type is stored in advance in association with identification information for each form type, and the distortion identifying unit stores the extracted distortion feature on the basis of the identification information. The distortion pattern is selected by comparing with the DB distortion pattern.
<Configuration 10>
A financial branch terminal equipped with the above character recognition device.

  By measuring the form distortion characteristics from the outline information of the form on the acquired image and comparing it with the registered distortion pattern, you can identify the distortion pattern of the form and perform character recognition on the part with less distortion .

  Examples of the present invention will be described as follows.

<Configuration of Example 1>
An overview of this system is shown in FIG.
This system includes a non-contact scanner device 201, an optical character recognition device 202, and a WS 203. The WS 203 includes an input unit such as a keyboard and a display unit such as a CRT. Hereinafter, description of the same parts as those of the conventional system will be omitted.

<Configuration of non-contact scanner device>
The configuration of the non-contact scanner device 201 is the same as that of the conventional system.

<Configuration of optical character recognition device>
As shown in FIG. 1, the internal configuration of the optical character recognition device 202 includes a control unit 202a, an image memory 202b, a form angle detection unit 202c, a format creation unit 202d, a format reading unit 202e, a character area correction unit 202f, a character recognition. 202g, side division unit 202h, form distortion feature extraction part 202i, form distortion identification part 202j, form distortion feature DB creation part 202k, form distortion feature DB reading part 202l, form distortion feature DB 202m, and allowable form distortion feature DB 202n. Is done.

  Since the control unit 202a, the image memory 202b, the form angle detection unit 202c, and the character recognition unit 202g are the same as those in the related art, description thereof will be omitted.

  The format creation unit 202d further includes a distortion feature registration unit 202d4 for registering reference distortion feature information serving as a criterion for distortion determination with respect to the configuration of the conventional system. The format creation unit 202d will be described later.

  The format reading unit 202e is a processing unit that provides a function of reading the contents of the format storage unit as in the conventional system, but in this embodiment, the reference distortion feature registered in the format storage unit is further added. Has the function to read and output.

The character area correction unit 202f is different from the conventional system in that the form is divided into four parts, upper left, upper right, lower right, and lower left, and upper left corner correction for correcting the character area included in each part. , An upper right end correction unit, a right lower end correction unit, and a left lower end correction unit.
That is, the upper left end correction unit has means for correcting the character region from the upper left form end coordinate. The upper right edge correction unit includes means for correcting the character area from the upper right form edge coordinates. The lower right end correction unit has means for correcting the character area from the lower right form edge coordinates. The lower left end correction unit has means for correcting the character area from the lower left form edge coordinates.

  The form distortion identification unit 202j uses the format reading unit 202e to acquire reference distortion feature information from the format storage unit, and compares the distortion characteristic information extracted by the form distortion feature extraction unit 202i to determine distortion of the form. Then, a corresponding distortion pattern is selected with reference to a later-described form distortion feature DB 202m, and whether or not the distortion pattern is allowable in the form is determined with reference to an allowable form distortion feature DB 202n described later. Provide functionality.

  The side dividing unit 202h provides a function for obtaining a side dividing point that is a reference for extracting each distortion feature by dividing each side of the form part under a predetermined condition. The side dividing unit 202h will be described later.

  The form distortion feature extraction unit 202i provides a function of extracting a form distortion feature from an image. The form distortion feature extraction unit 202i will be described later.

<Distortion feature information>
Here, the distortion feature information in the present embodiment will be described.
FIG. 5A shows a state in which distortion features are extracted from a form portion of a form image that is a basis for distortion information. Note that the density of the halftone dots is different from that in FIGS. 12 (a) and 12 (b), but this is for easy viewing of the characters, and the difference in density has no particular meaning.

In FIG. 5A, each symbol has the following meaning.
Te: Width of the upper side of the form part Be: Width of the lower side of the form part Le: Vertical width of the left side of the form part Re: Vertical width of the right side of the form part n: Number of divisions of the upper side and lower side m: Number of divisions of the left side and right side Vex (x = 1, 2,... N): Line segment connecting the upper side and lower side opposing side division points Hey (y = 1, 2,... M): Line segment connecting the left side and right side opposing side division points θeyx (y = 1, 2,... m) (x = 1, 2,... n): Angle of intersection of Vex and Hey Each extraction method will be described later.

  In the present embodiment, θeyx is extracted as distortion feature information using one reference form, and is stored in the format storage unit as common reference distortion feature information.

FIG. 5B shows a state in which distortion features are extracted from the form portion of the form image to be determined by the same process as in FIG.
FIG. 5B emphasizes and describes the distortion of the form when an image is acquired with the left corner of the form raised. The meaning of the symbol is the same as that in FIG. 5A, only that e is changed to r.

<Format creation section>
As shown in FIG. 3, the internal configuration of the format creation unit 202e includes a control unit 202d1, a format storage unit 202d2, a character area coordinate setting unit 202d3, and a distortion feature registration unit 202d4. Moreover, it has a communication processing unit (not shown) with other processing units.

  Here, the distortion feature registration unit 202d4 added to the conventional system will be described.

  The distortion feature registration unit 202d4 provides a function of registering, in the format storage unit, distortion feature information extracted for the form portion of the form image of the reference form. For example, when an operator gives a start instruction to the format creation unit via the WS 103, a series of processing from extraction of distortion feature information to registration in the format storage unit is performed.

  The format only needs to be registered in advance, and it is sufficient that the format registered in the format storage unit can be referred to during operation. Therefore, the format creation unit 202d can be omitted during operation.

  Also, an allowable distortion generation area is registered according to information on where the character recognition area is located in the upper left, upper right, lower left, and lower right. This may be performed automatically depending on the position where the character recognition area exists, or may be specified by an operator. This information is registered in the allowable form distortion feature DB 202n as shown in FIG.

<Side division part>
The side dividing unit 202h provides a function of dividing the sides of the form part.
Hereinafter, processing when the side is divided by the side dividing unit 202h will be described.

First, the corner of the form portion is detected by the corner detection unit 202c as in the conventional system.
The side dividing unit 202h obtains a line segment connecting the detected corners. Subsequently, each of the obtained line segments is divided into equal intervals by a predetermined number. The predetermined numbers are n and m in FIGS. 5A and 5B, and the number of division points is constant regardless of the size of the input form. Subsequently, for example, if the upper side and the lower side are scanned, the pixel is scanned in the Y-axis direction (if the left side and the right side are the x-axis direction), and the points where the background color and the density of the form portion change are extracted. By doing so, it is possible to extract the edge dividing points of the respective document edges corresponding to the dividing points of the line segments.

<Form distortion feature extraction unit>
The form distortion feature extraction unit 202i obtains a line segment that connects the opposing points as shown in the drawing with respect to the side division points obtained by the side division unit 202h. These line segments are like Vex, Hey, Vrx, and Hry in FIGS. Subsequently, the angle θeyx of each intersection is obtained and output as a feature.

Note that the presence / absence of form distortion is evaluated as follows. For example, the equation for evaluating the angle is as follows.
[Form distortion evaluation value]
θyx = | θeyx_θryx | (y = 1, 2... m) (x = 1, 2... n)
[Threshold of form distortion evaluation value]
The threshold value of θyx: θTHr

<Form distortion feature DB>
An example of the assumed distortion pattern is shown in FIG. The shaded portion is a portion where distortion occurs, that is, a portion where a predetermined threshold is exceeded, and a portion where distortion is large. A portion without shading is a portion where distortion is small or not generated, that is, a portion having a predetermined threshold value or less.

The form distortion feature DB 202m stores, as shown in FIG. 13, the name, code, and distortion pattern condition of each distortion pattern as the form distortion feature corresponding to the assumed distortion pattern shown in FIG. For example, in the case of “lifting the upper left corner”, the distortion pattern condition corresponds to a case where only the upper left portion of the above-described θeyx exceeds the threshold. For example, when the threshold value is exceeded only for nine points θe11 to θe33 among all θeyx, it can be determined that the upper left corner is lifted (distortion pattern code = A).
Further, for example, in the case of “left side & right side floating”, only the both end portions of Vex described above are longer than the other, and the ratio exceeds a predetermined threshold value. For example, when only two of Ve1 and Ven exceed the average of the lengths of Vex and are longer than the threshold, it is possible to determine that “left side & right side floating” (distortion pattern code = U).
The distortion pattern condition describes the relationship between how to float and the above-described Vex, Hey, and θeyx. Actually, the sentences in the table are expressed and recorded as logical expressions.

In addition, information on which one of the upper left correction unit, the upper right correction unit, the lower right correction unit, and the lower left correction unit can be used is included. For example, in the case of FIG. 13, in the case of “lifting the upper left corner”, only the upper left end correction unit is unusable, and the rest is usable. Further, in the case of “left side & right side floating”, it is considered that distortion occurs at all corners, so that all the correction units cannot be used.
The form distortion feature DB is registered by the form distortion feature DB creation unit 202k as described later.

<Allowable form distortion feature DB>
The allowable form distortion feature DB 202n is a table in which distortion pattern codes of allowable distortion patterns are registered for each form whose format is registered as a processing target. In the example of FIG. 4, it can be seen that the form distortion pattern allowed by the form 001 (form ID = 001) is A (upper left corner). For example, a case in which an explanatory note for the writer is printed in the upper left part and a character recognition area exists in the other part corresponds to the case.

<Form distortion feature DB creation section>
The form distortion feature DB creation unit 202k is a processing unit for registering the form distortion feature DB 202n. Specifically, an image corresponding to an assumed form distortion pattern as shown in FIG. 14 is intentionally created and read, and a form distortion feature for each form distortion pattern is generated and registered. Also, in each form distortion pattern, information indicating which of the correction units of the upper left end correction unit, the upper right end correction unit, the lower right end correction unit, and the lower left end correction unit can be used is set according to the degree of occurrence of distortion. The For example, if the distortion in the upper left region is less than a predetermined threshold, the upper left correction unit can be used. The same applies to other regions. Note that if the distortion is greater than or equal to a predetermined threshold value for all regions, all the correction units may not be usable.

<Operation of Example 1>
(A) Format Creation FIG. 15A is a flowchart at the time of registration of the reference distortion feature information in the optical character recognition system using a non-contact scanner. First, the image acquisition unit 201b reads a reference registered form placed on the mat unit 201a, and takes the acquired image into the image memory 202b (S301). Here, since it is necessary to acquire the image of the reference form in a state without distortion, it is necessary to place the image in a state in which there is no lifting or bending. This is called a standard state or a reference state.

  Subsequently, the form angle detecting unit 202c detects four form angles on the form image (S302).

  Subsequently, the side dividing unit 202h divides the sides of the form portion to obtain side dividing points (S303).

  Subsequently, the form feature extraction unit 202i extracts distortion feature information of the reference form image (S304).

  Subsequently, the distortion feature registration unit 202d4 registers the extracted distortion feature information in the format storage unit as reference distortion feature information. (S305).

  Note that all the angles included in the reference distortion feature information acquired in the standard state are considered to be approximately 90 degrees, so that all the angles are set to 90 degrees without acquiring the distortion characteristics in the standard state as described above. It is also possible to generate and register distortion feature information. However, if the distortion feature information in the standard state is acquired and the distortion feature information is actually extracted as described above, it is possible to eliminate the effects of distortion occurring in the standard state due to the characteristics of the camera lens and the like. , Accuracy can be improved.

  Further, the operator designates a character recognition area for each form based on the acquired reference form image and registers it in the format. As for the information of the designated character recognition area, information on the relative coordinates from the reference angle and information on the area size are registered as coordinate information as well as the coordinate information of the reference corner. After that, the format is saved. (S306)

(B) When creating a form distortion feature DB FIG. 15B is a flowchart when creating the form distortion feature DB in the optical character recognition system using a non-contact scanner. In this embodiment, it is assumed that the form distortion feature DB is common to all formats. For this reason, it is desirable to create the form distortion feature DB by using the form used for registering the reference distortion feature information before the format creation shown in FIG.

  First, a form is placed on the mat part 201a in a state where it floats. The image acquisition unit 201b reads the registration form placed on the mat unit 201a, and takes the acquired image into the image memory 202b (S307).

  Subsequently, the form angle detection unit 202c detects four form angles on the form image (S308).

  Subsequently, the form feature extraction unit 202i extracts the distortion feature information of the form image (1309).

  Subsequently, the form distortion feature DB creation unit 202k learns the form distortion feature and creates the form distortion feature DB (S310).

  The operations of S307 to S310 are performed for all assumed form distortion patterns (examples are shown in FIG. 14). (S311)

(C) Reading time FIG. 16 is a flowchart at the time of reading in the optical character recognition system using the non-contact scanner. First, the image acquisition unit 201b reads a read form placed on the mat unit 201a, and takes the acquired image into the image memory 202b (S401).

  Subsequently, the form angle detection unit 202c refers to the image memory 202b and detects four form angles on the image (S402).

  Subsequently, the side dividing unit 202h divides the sides of the form portion and obtains side dividing points (S403).

  Subsequently, the form distortion feature extraction unit 202i extracts the form distortion feature of the form part of the image (S404).

  Subsequently, the format reading unit 202e acquires reference distortion feature information from the format storage unit. (S405)

  Subsequently, the form distortion identification unit 202j calculates a form distortion evaluation value shown below, and compares it with each preset threshold (θTHr) to detect a form distortion feature. At this time, the form distortion feature DB reading unit 202l refers to the form distortion feature DB 202m and outputs a form distortion pattern code as an identification result if the corresponding form distortion pattern exists. If no corresponding form distortion pattern is found, the fact that identification has failed is output. (S406)

  If identification of the form distortion feature has failed (S407), “distortion detection failure” is displayed on the CRT of the WS 203 (S408), and “form reject” is further displayed on the CRT of the WS 203 (S409).

When the identification is successful and the identification result is not distorted (S410), the format reading unit 202e reads the coordinate information of the reference angle of the form and the coordinate information of the character recognition area. (S411).
Subsequently, as in the prior art, the character area is corrected using the reference corner coordinates (for example, the upper left corner coordinates) (S412). Subsequently, the character recognition unit 202g recognizes the character (S413), and displays the character recognition result on the CRT of the WS 203 (S414).

  When the identification result is a correctable distortion feature pattern, the format reading unit 1006 reads the coordinate information of the reference angle of the form and the coordinate information of the character recognition area by the format reading unit 202e. (S415). Subsequently, according to the identification result, the form corner coordinates / character region coordinates of the registered image are appropriately used by using the upper left end correction unit, upper right end correction unit, lower right end correction unit, and lower left end correction unit of the character region correction unit 202f. The character area is corrected from the form corner coordinates of the read image (S416).

For example, FIG. 17 shows an example of correction in the case where the identification result is a lift in the upper left corner.
As shown in the figure, since the character area near the upper left corner cannot be corrected, the character area near the upper left corner is rejected. That is, the character recognition result is unread. For a character region far from the upper left corner, the character region is corrected from angular coordinates other than the upper left corner. That is, the form is divided into four parts, upper left, upper right, lower right, and lower left, and corrected by the above-described upper left end correction unit, upper right end correction unit, right lower end correction unit, and left lower end correction unit, respectively.
Subsequently, in the character area that is not rejected, the character recognition unit 202g recognizes the character (S417), and the character recognition result is displayed on the CRT of the WS 203 (S418). Subsequently, the form distortion identification result is displayed on the CRT of the WS 203 (S419). If the identification result is an uncorrectable distortion feature pattern (S410), the form distortion identification result is displayed on the CRT of the WS 203 (S420), and “form reject” is displayed on the CRT of the WS 203 (S409).

  In this way, each side of the reference form image is divided by a predetermined value to obtain a side division point, and the angle of the intersection of the line segments connecting the respective sides is set as a reference distortion feature, which is a determination target when determining distortion. Each side of the form image is divided in the same way as the standard form, and the angle of the intersection of the line segments that connect the same is used as the distortion feature. By comparing with the standard distortion feature, the distortion on the form image is identified, and the character Character recognition is performed on the recognizable area and the result of identification of the form distortion is displayed, so that the operator can be warned, the occurrence of financial transaction accidents due to misalignment of the character area can be prevented, and the burden of confirmation also for the operator It is possible to obtain an effect that decreases.

  Further, regarding the assumed form distortion, the form distortion DB is automatically created by reading the image in which the distortion is actually generated, so that it is possible to reduce the impossibility of the operator's registration work.

  In addition, since only the angle of the intersection of the line segments connecting the side division points is characterized, the same number of intersections can be obtained even if the non-decision form is different in size from the reference form. Since common feature data can be used, registration is required only once, and image distortion including new forms can be determined.

  In the first embodiment, only the angle of the intersection of the line segments connecting the corresponding edge division points is used as the distortion feature. However, in this embodiment, the length of the line segment is also used as the distortion feature information. For this reason, since the reference distortion feature differs for each type of form, the reference distortion feature information is registered for each format of each form. The form distortion feature DB is also registered as information for each form.

<Configuration of Example 2>
The system configuration in the present embodiment zero is shown in FIG. Only the parts different from the first embodiment will be described below.

<Configuration of optical character recognition device>
The internal configuration of the optical character recognition device 202 is shown in FIG.
Since the control unit 202a, the image memory 202b, the form angle detection unit 202c, the character area correction unit 202f, and the character recognition unit 202g are the same as those in the related art, description thereof will be omitted.

  In the present embodiment, the format reading unit 202e has a function of acquiring the reference distortion feature for each format.

  The form distortion feature extraction unit 202i provides a function of extracting a form distortion characteristic from an image in the same manner as in the first embodiment, but in addition to the angle of the intersection of the line segments connecting the side division points, The length of the minutes is also extracted as feature information. The form distortion feature extraction unit 202i will be described later.

<Distortion feature information>
Here, the distortion feature information in the present embodiment will be described.
FIG. 5A shows a state in which distortion features are extracted from a form portion of a form image that is a basis for distortion information. The meaning of each symbol is the same as in the first embodiment.

In this embodiment, using a single reference form, Vex, Hey, and θeyx are extracted as distortion feature information and stored in the format as common reference distortion feature information. Here, unlike the first embodiment, the reference distortion feature needs to be registered for each type of form registered in the format.
Since FIG. 5B is the same as that of the first embodiment, description thereof is omitted.

<Format creation section>
The internal configuration of the format creation unit 202e is shown in FIG.
The distortion feature registration unit 202d4 stores the reference distortion feature in the format storage unit in association with the format of each form. For this reason, it is efficient to perform the reference distortion feature information together with the registration of the character information for each form.

<Form distortion feature extraction unit>
The form distortion feature extraction unit 202i obtains a line segment that connects opposite sides of the side division points obtained by the side division unit 202h. These line segments are like Vex, Hey, Vrx, and Hry in FIGS. Subsequently, an angle θeyx of each intersection is obtained and output as a feature together with the length of each line segment.

<Form distortion feature DB creation section>
The form distortion feature DB creation unit 202k registers the form distortion feature DB 202m for each form ID of each form, not in common. For this reason, the reference distortion feature information is performed using the form together with the registration of the character information for each form.

<Operation of Example 2>
(A) At the time of format creation The operation in this embodiment will be described with reference to FIG. About S301-S304, since it is the same as that of Example 1, description is abbreviate | omitted.

  In S305, the distortion feature registration unit 202d4 registers the extracted distortion feature information as reference distortion feature information in association with the format of the form.

(B) At the time of creating the form distortion feature DB The operation in the present embodiment will be described with reference to FIG. In the present embodiment, the form distortion feature DB is created for each form. That is, a table as shown in FIG. 13 is individually registered for each form ID. Since other than that is the same as that of Example 1, description is abbreviate | omitted.

(C) Reading The operation in the present embodiment will be described with reference to FIG. Since steps other than S405 and S406 are the same as those in the first embodiment, description thereof is omitted.

  In step S405, the format reading unit 202e acquires reference distortion feature information corresponding to the form type of the form image from the format storage unit (405). An existing method is applied to specify the type of form. For example, feature information such as sheet ID and ruled line feature of the form is registered together with the format, and a format that matches the feature is selected. Good. Alternatively, a form type designation means may be provided to the WS operator, and the form type may be specified based on the designated information.

In S406, the form distortion identification unit 202j calculates a form distortion evaluation value using the following formula.
[Form distortion evaluation value]
Vx = | Vex−Vrx | (x = 1, 2,... N)
Hy = | Hey−Hry | (y = 1, 2,... M)
θyx = | θeyx−θryx | (y = 1, 2,... m) (x = 1, 2,... n)
[Threshold of form distortion evaluation value]
Hx threshold: VTHr
Hy threshold: HTHr
The threshold value of θyx: θTHr
The form distortion feature is detected by comparing with respective threshold values (VTHr, HTHr, θTHr) set in advance. Then, the form distortion feature DB reading unit 202l refers to the form distortion feature DB 202m of the form and outputs a form distortion pattern code as an identification result if the corresponding form distortion pattern exists. If no corresponding form distortion pattern is found, the fact that identification has failed is output. (S406)

  In this way, each side of the reference form image is divided by a predetermined value to obtain a side division point, and the angle of the intersection of the line segment connecting the respective sides and the length of each line segment are registered for each form as a reference distortion feature. When determining distortion, each side of the form image to be determined is divided in the same manner as the reference form, and the angle of the intersection of the line segments connecting the respective lines and the length of the line segment are also distortion characteristics. Since the distortion on the image of the form can be determined by comparing with the feature, the distortion can be determined with higher accuracy than the first embodiment.

In the above-described embodiment, the line connecting the corners is equally divided. However, since the distortion due to the lift of the form is closer to the corner of the form, the division interval of the part near the corner of the form is larger. By shortening and increasing the division interval of the center, that is, the part far from the corner, the distortion of the form part can be detected with higher accuracy when the density of the side division points is given. For example, by dividing the division intervals corresponding to n and m in FIGS. 5A and 5B using a trigonometric function, it is possible to shorten the division interval near the corner. For example, if the center of the side is set as the starting point = 0 and the corner of the side is set as the end point = π / 2 [rad], the points obtained by equally dividing the side can be mapped with a sine function so that the intervals become shorter as the angle becomes closer. For example, when the center to the corner is divided into three, the point 1/3 from the center is sin ((π / 2) × (1/3)) = sin (π / 6) = 0.5, so Maps to half the point to the corner. The point 2/3 from the center is approximately 0.87 times from the center to the corner by sin ((π / 2) x (2/3)) = sin (π / 3) = about 0.87. Mapped.
In the first embodiment, the common reference distortion feature information may be registered once. However, the reference distortion feature information may be registered for each form by the method as in the second embodiment. For example, when there is a difference in distortion depending on the size of the form due to the characteristics of the camera lens or the like, the accuracy can be further improved by registering the reference distortion feature information for each form.

In the above-described embodiment, the optical character recognition device 102 and the scanner device 201 have been described as different configurations. However, the scanner and the optical character recognition device may be configured as a single body.
Further, by combining such optical character recognition device 102 and WS 203 as described above, it is possible to function as a terminal device having a character recognition function for a form image. In particular, when WS 203 is a financial sales office window terminal, it can function as a financial sales office window terminal having a function for recognizing a form image.

Functional block of optical character recognition device of the present invention System configuration of the present invention Functional block of format creation unit of the present invention Permissible form distortion feature DB for each form Explanation of distortion characteristics of forms Conventional system configuration Functional block of conventional scanner Functional block of conventional optical character recognition device Functional block of conventional format creation unit Functional block of conventional character area correction unit Conventional flowchart Acquired image and area correction result at the time of form distortion Report distortion feature DB Example of distortion pattern Flow chart of the present invention Flow chart of the present invention Acquired image and area correction result at the time of form distortion

Explanation of symbols

201 Scanner Device 201a Matte Unit 201b Image Acquisition Unit 202 Optical Character Recognition Device 203 WS
202a Control section 202b Image memory 202c Form angle detection section 202d Format creation section 202e Format reading section 202f Character area correction section 202g Character recognition section 202h Side division section 202i Form distortion feature extraction section 202j Form distortion identification section 202k Form distortion feature DB creation section 202l Form distortion feature DB reading unit 202m Form distortion feature DB
202n Allowable form distortion feature DB

Claims (10)

A character recognition device that acquires an image of a form and performs character recognition in accordance with distortion of the outer shape of the form part on the image,
A distortion feature extraction unit for extracting distortion characteristics of the outer shape of the form part;
A form distortion feature DB that classifies the distortion patterns according to the degree of distortion for each part of the form;
A distortion identifying unit for selecting a corresponding distortion pattern based on the extracted distortion characteristics of the form part;
A character recognition unit that performs character recognition according to the selected distortion pattern,
Character recognition device.   The character recognition device according to claim 1, wherein the character recognition unit performs character recognition on a portion with a small distortion based on the selected distortion pattern. The character recognition unit selects a character region coordinate correction method according to a distortion pattern;
The character recognition device according to claim 1. An edge dividing unit for obtaining an edge dividing point obtained by dividing an edge of the form part,
The distortion feature extraction unit connects side division points of the opposing sides according to a predetermined correspondence condition, and extracts an angle between line segments at the intersection of each line segment as a distortion feature.
The character recognition device according to claim 1. A corner detection unit for detecting a corner of the form part;
The character recognition device according to claim 4, wherein the side dividing unit obtains points obtained by dividing the lines connecting the detected corners, and divides the sides by associating the obtained points with the sides of the form part.   The character recognition device according to claim 5, wherein the side dividing unit divides the division into a predetermined number of equal parts.   The character recognition device according to claim 5, wherein the side dividing unit shortens an interval of the division in a portion close to a corner.   The character recognition device according to claim 7, wherein the division interval is determined by mapping a position of a division point when equally divided by a trigonometric function. In the form distortion feature DB, a distortion pattern for each form type is stored in advance in association with identification information for each form type,
The character recognition according to claim 1, wherein the distortion identification unit selects the distortion pattern by comparing the extracted distortion feature with a distortion pattern of a form distortion feature DB based on the identification information. apparatus.   A financial store terminal comprising the character recognition device according to claim 1.

Images