JP4836826B2 - Image processing method of character string image, program thereof and apparatus thereof - Google Patents

Description

  The present invention relates to a character string image image processing method for specifying an existing area of each character image based on an image obtained by imaging a member on which a plurality of character string images are formed, a program therefor, and a program therefor It relates to the device.

At the manufacturing and distribution sites, character strings (hereinafter referred to as detection target character strings) representing management information such as lot numbers and manufacturing numbers are printed on the surface of managed members such as products and intermediate products. A management system is employed that performs image processing for character recognition by a computer based on an image obtained by imaging with a camera, and manages a managed member based on the result of the character recognition. In such a management system, the character string to be detected is directly printed on the surface of the managed member by a printer head, or the character string to be detected is printed on a predetermined seal member, and the seal member is used as the managed member. Or affixed to
By the way, in the production and distribution sites, the image of the managed member is not good because the environment of printing and imaging is not good, such as movement and dirt of the managed member, contact with other members, insufficient lighting, etc. The quality of the image of the detection target character string may be poor. For example, in the image of the character string to be detected, there may be a positional shift between a plurality of character strings, bleeding of character images, partial loss (disturbance of character images), or the like.
FIG. 12 is an example of a binary image with poor image quality for some character images. The example shown in FIG. 12 is an example in which the quality of the character string image of one column out of the images of the character string to be detected consisting of two character strings is in a poor state due to dirt or the like.
In order to manage managed members correctly, even if the quality of the detection target character string in the captured image is poor, each character image in the captured image (the image of each character constituting the detection target character string). ) Is correctly specified, and character recognition by pattern recognition or the like is required for the image (character image) in that region.
On the other hand, for example, in Patent Document 1, a character candidate image that is a continuous pixel satisfying a predetermined condition is extracted from a character string image, and the arrangement between the character candidates is based on the character candidate image. A method for determining a character cutout region (character region) from a circumscribed rectangle is shown.
In Patent Document 2, a character string image size (character height) is obtained based on projection data in the character string direction, and further, based on projection data in a direction orthogonal to the character string direction. Thus, a method is described in which the line size (character width) is obtained and the character cutout position is specified based on the column size and the line size thus obtained.
JP-A-8-339421 Japanese Patent Laid-Open No. 10-69523

By the way, the image of the detection target character string composed of a plurality of character strings is printed on the managed member, and there may be a large positional shift in the column direction (character string direction) between the images of the character string for one column. . For example, when a print head for printing the character string to be detected is provided for each character string, or when a character string for one row is printed individually in each of a plurality of processes, etc. is there.
In addition, a gap (a non-image area) between the character image and the character image is clear due to problems such as dirt, poor lighting, and contact with other members of a part of the characters constituting the detection target character string. It may not be. For example, a certain character image is connected to another adjacent character image (hereinafter referred to as a connected state), or a character image that should originally be one connected pixel is broken into a plurality of connected pixels ( A typical example is a case where it is referred to as a “ruptured state” below.
Even in such a situation where the quality of the image of the character string to be detected is poor, it is desirable that the existing area of each character image can be specified as accurately as possible.
However, in the technique disclosed in Patent Document 1, in the entire image of the character string to be detected, the circumscribed rectangles of each character image (the circumscribed rectangle of the connected pixels) are substantially deviated from the predetermined positional relationship. It is assumed that there is no. For this reason, in the case where there is a large positional shift in the column direction between the images of each of the character strings in a plurality of columns of the detection target character strings, the technique disclosed in Patent Literature 1 cannot correctly identify the existence region of each character image. There was a problem.
In addition, for a part of characters constituting the detection target character string, when the character image is in the connected state or the broken state, that is, one connected pixel and one character image are in a one-to-one relationship. If not, the positional relationship between the circumscribed rectangles of the connected pixels does not satisfy a predetermined positional relationship, and the technique disclosed in Patent Document 1 still cannot correctly identify the existence area of each character image. there were.
On the other hand, the technique disclosed in Patent Document 2 specifies the existence range of each character string based on the projection data in the column direction in the image of the detection target character string, and then determines the range of the character string for one column. For each image, the area of each character image is specified based on projection data in a direction orthogonal to the column direction. For this reason, according to the technique shown in Patent Document 2, even if the positional deviation in the column direction between the images of each of the character strings for one column in the plurality of character strings to be detected is relatively large, Each existence area can be correctly identified.
However, when a gap between a character image and a character image is not clear for a part of characters constituting the detection target character string, a column including such a character image is obtained by the technique disclosed in Patent Document 2. Even if the character area is specified based on the projection data orthogonal to the column direction, there is a problem that the existence area of each character image cannot be specified correctly.
Accordingly, the present invention has been made in view of the above circumstances, and the object of the present invention is to identify one character image area when specifying an area of each character image with respect to an image of a character string to be detected consisting of a plurality of character strings. Even when there is a large misalignment in the column direction between the images of the character strings, or when the gap between the character images is not clear for some of the characters constituting the detection target character string It is an object of the present invention to provide a character string image image processing method, a program thereof, and an apparatus thereof that can specify the existence region of each character image as accurately as possible.

In order to achieve the above object, an image processing method for a character string image according to the present invention is composed of a plurality of character strings, and the character size, the number of characters in one column, the character spacing, the number of columns, and the column spacing are predetermined (known. ) Reads out binary image data of an image obtained by imaging a member on which an image of a detection target character string that satisfies the character arrangement condition is formed from a predetermined storage unit, and based on the read data, the binary image Is a method for identifying an existing area of each character image constituting the character string to be detected and extracting a character image from the identified existing image area of each character, and a predetermined processor (may be referred to as a computer) Thus, the following procedures (1) to (9) are automatically executed. The character image in the detection target character string to be detected in the present invention is originally an image composed of connected pixels when there is no deterioration in image quality due to poor printing, dirt, poor lighting, or the like.
(1) A first character candidate specifying procedure for specifying, based on the read data, a connected pixel in the binary image that satisfies a predetermined character image condition.
(2) Among the connected pixels specified by the first character candidate specifying procedure, the relative positional relationship satisfies the relative positional relationship between characters in the character arrangement condition within a predetermined error range. A second character candidate identification procedure for identifying.
(3) Based on the position and size of each character candidate image that is the connected pixel specified by the second character candidate specifying procedure, the character arrangement condition, and a predetermined character arrangement variation range, A character string inclusion region estimation procedure for estimating a region including all of the character candidate images in the value image and including the entire image of the detection target character string.
(4) By performing a predetermined process on the image of the character string inclusion estimation area, which is an area estimated by the character string inclusion area estimation procedure in the binary image, a single line constituting the detection target character string The column width range estimation procedure which estimates the existence range in the column orthogonal direction orthogonal to the column direction of each character string image.
(5) Correction processing for limiting the upper limit value of projection data in the column orthogonal direction or projection data in the column orthogonal direction with respect to a row region image that is an image of each range estimated by the row width range estimation procedure in the character string inclusion region image A first column orthogonal direction projection data calculation procedure for calculating the distribution of the data subjected to.
(6) A cross-correlation function derivation procedure for deriving a cross-correlation function with the shift amount in the column direction as a variable based on calculation data obtained by the first column orthogonal direction projection data calculation procedure.
(7) A column mutual displacement amount derivation procedure for deriving a column mutual displacement amount, which is a positional displacement amount in the column direction between the column region images, based on the cross correlation function obtained by the cross correlation function derivation procedure. (8) Calculate projection data in the column orthogonal direction in the entire image obtained by correcting the position of each of the plurality of column region images by the column mutual shift amount or a distribution of data subjected to correction processing for limiting the upper limit value of the projection data. A second column orthogonal direction projection data calculation procedure.
(9) Character area specification that specifies the position of the existence area of the image of each character constituting the detection target character string in each of the row area images based on the calculation data obtained by the second column orthogonal direction projection data calculation procedure procedure.
The predetermined character image conditions include, for example, the size of the circumscribed rectangle of the connected pixel, the aspect ratio of the circumscribed rectangle of the connected pixel, the area of the connected pixel, the area ratio of the connected pixel and its circumscribed rectangle, and That is, at least one of the ratio of the square of the perimeter of the connected pixel and its area is within a predetermined range.
Further, for example, the processor, in the column mutual displacement amount guiding out procedure, the columns each other on the basis of the shift amount in the column direction corresponding to the maximum correlation value in said cross-correlation function obtained by the cross-correlation function derived Procedure It is conceivable to derive the deviation amount.

In the present invention, the processor executes, for example, the following procedures (10) and (11) in the column width range estimation procedure.
(10) A column direction projection data calculation procedure for calculating a distribution of projection data in the column direction for the image of the character string inclusion estimation region in the binary image.
(11) A range estimation procedure for estimating an existence range in the column orthogonal direction of each character string image constituting the detection target character string based on calculation data obtained by the column direction projection data calculation procedure.
An example of the processing according to the procedures shown in (10) and (11) is processing in the step of cutting out the character string shown in Patent Document 1.
In addition, when there is little need to consider the positional deviation in the column orthogonal direction of the image of the character string for one column, the processor equally divides the character string inclusion estimation area by the number of columns of known character strings. Thus, it is conceivable to estimate the existence range in the column orthogonal direction of each character string image for one column.
In the present invention, it is conceivable that the processor further automatically executes the following procedures (12) and (13).
(12) The detection target in each of the row region images based on the interval between peaks of a plurality of correlation values in the cross-correlation function obtained by the cross-correlation function derivation procedure and the number of characters in one row in the character arrangement condition A column length calculation procedure for calculating a length in a column direction of a range where a character string image exists.
(13) With respect to calculation data obtained by the second column orthogonal direction projection data calculation procedure, a value within a column length range that is a calculation result obtained by the column length calculation procedure with respect to each of a plurality of reference positions in the column direction. A column region image processing target range specifying procedure for calculating an integrated value or an average value and specifying a processing target range in the column region image based on the reference position and the column length when the calculated value is maximized.
In this case, the processor, in the character region specifying procedure, includes the presence of an image of each character constituting the detection target character string for the processing target range specified by the column region image processing target range specifying procedure. Identify the area.

The present invention is embodied as a character image image processing program for causing a predetermined processor (may be referred to as a computer) to execute each procedure in the character image image processing method according to the present invention described above. It is also possible.
Similarly, the present invention may be embodied as a character image image processing apparatus including means for executing each procedure in the character image image processing method according to the present invention described above. A computer (information processing apparatus) that executes each procedure in the above-described character image image processing method according to the present invention by executing a predetermined program is used in the character image image processing apparatus according to the present invention. This is a typical example.

  According to the present invention, as will be described later, when specifying the area of each character image with respect to the image of the character string to be detected consisting of a plurality of character strings, the positions of the character strings for one column in the column direction between the images. Even if the deviation is large or the gap between the character images is not clear for a part of the characters constituting the detection target character string, the existence area of each character image can be specified as accurately as possible. .

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings so that the present invention can be understood. The following embodiment is an example embodying the present invention, and does not limit the technical scope of the present invention.
FIG. 1 is a block diagram showing a schematic configuration of a character image image processing apparatus Z according to an embodiment of the present invention. FIG. 2 is a flowchart showing a procedure of character image cutting and character recognition processing by the image processing apparatus Z. 3 is a diagram illustrating an example of a binary image to be subjected to image processing by the image processing device Z, FIG. 4 is a diagram illustrating an example of a character image candidate in the binary image, and FIG. 5 is a character arrangement condition in a detection target character string FIG. 6 is a diagram showing the positional relationship between character image candidates in a binary image, and FIG. 7 is an example of an image of a character string inclusion estimation region specified by the image processing device Z. FIG. 8 is a diagram showing an example of a character string inclusion estimation region image and a projection data distribution in the column direction in the image, and FIG. 9 is a column region image in the character string inclusion estimation region and projection data in the column orthogonal direction in the image. distribution FIG. 10 is a graph showing an example of a cross-correlation function of data obtained by correcting projection data in the column orthogonal direction of the row region image. FIG. 11 is an image in which the positional deviation in the row direction of the row region image is corrected, and its FIG. 12 is a diagram illustrating an example of the distribution of data obtained by correcting the upper limit of projection data in the column orthogonal direction of the entire image, and FIG. 12 is a diagram illustrating an example of a binary image with poor image quality for some character images.

First, the configuration of a character image image processing apparatus Z (hereinafter referred to as an image processing apparatus Z) according to an embodiment of the present invention will be described with reference to the block diagram shown in FIG.
The image processing device Z is a computer and includes a CPU 41, a camera interface 42, an image memory 43, an image processing unit 44, a display unit 45, a general purpose memory 46, and a communication unit 47, as shown in FIG.
The image processing device Z stores binary image data based on image data obtained by imaging the managed member 1 on which the image c of the detection target character string is formed by the camera 3 (imaging means). 43, and based on the read data (binary image data), the existence region of each character image constituting the detection target character string in the binary image is specified, and the existence of the specified character image A device that extracts a character image from a region and automatically executes a character recognition process on the extracted character image.

The camera interface 42 controls the camera 3 that captures the managed member 1 on which an image c of a predetermined character string to be detected is printed. Specifically, the camera interface 42 controls the shutter of the camera 3 to be released when the managed member 1 conveyed by the conveying means 2 such as a belt conveyor reaches the vicinity of the front of the camera 3. At the same time, data of an image captured by the camera 3 is acquired from the camera 3 and stored in the image memory 43.
The managed member 1 is, for example, a slab or billet, and an image c of the detection target character string is printed on the surface thereof.
The detection target character string is composed of a plurality of character strings, and the character size, the number of characters in one column, the character interval, the number of columns, and the column interval satisfy a predetermined character arrangement condition. In addition, the image of each character constituting the detection target character string is originally an image composed of connected pixels when there is no deterioration in image quality due to poor printing, dirt, poor lighting, or the like.
FIG. 5 is a diagram schematically showing the character arrangement conditions in the detection target character string.
The character string to be detected is composed of a plurality of character strings, and the height FH and width FW of the circumscribed rectangle of each character, the number of characters NC and character spacing dFW, the number of columns NL, and the column spacing dFH are known. FIG. 5 shows that the number of characters NC in one column is 3, and the number of columns NL is 2. Information on character arrangement conditions including information on these FH, FW, NC, dFW, NL, and dFH is stored in advance in the general-purpose memory 46, and the CPU 41 refers to the information on the character arrangement conditions as necessary. To do.

The CPU 41 is a processor that controls the camera interface 42, the image processing unit 44, the display unit 45, and the communication unit 47 by executing a program stored in advance in the general-purpose memory 46.
The image memory 43 is storage means for storing various image data such as data of an image captured by the camera 3 and data of its binary image.
The image processing unit 44 is a processor that executes a program stored in advance in a ROM (not shown). The image processing unit 44 reads image data from the image memory 43 and performs various image processing (processing of image data) on the read image data. At the same time, the image memory 43 stores the image data after the image processing.
The CPU 41 and the image processing unit 44 are an example of a processor that executes each procedure in the image processing method for character images according to the present invention. The program executed by each of the CPU 41 and the image processing unit 44 is an example of a character image image processing program according to the present invention.
The display unit 45 is a video display unit that displays an image based on the image data stored in the image memory 43 and displays information delivered from the CPU 41 as an image.
The general-purpose memory 46 is a non-volatile memory that stores various data read and written by the CPU 41 and programs executed by the CPU 41, and is, for example, a hard disk or a flash memory.
The communication unit 47 is a communication unit that communicates with a host computer that manages information related to the managed member 1, and management information obtained by character recognition processing based on the image c of the detection target character string (the detection target Information represented by a character string) is transmitted to the host computer.

Next, a procedure of character image cutout / character recognition processing executed by the image processing device Z (computer) will be described with reference to the flowchart shown in FIG. The following processing of the camera interface 42 and the image processing unit 44 is executed according to a control command from the CPU 41. S1, S2,... Shown below represent identification codes of processing procedures (steps). Further, the processing of steps S1 to S19 shown below is automatically executed by the CPU 41 and the image processing unit 44.
[Steps S1, S2]
First, when the CPU 41 detects that the managed member 1 has arrived near the front of the camera 3 based on a detection signal of a sensor (not shown), the camera interface 42 causes the camera 3 to perform imaging and the captured image of the captured image is displayed. Data is acquired from the camera 3, and the image data is stored in the image memory 43 (S1).
Further, the image processing unit 44 performs binarization processing on the captured image data stored in the image memory 43 in accordance with the control command of the CPU 41, and the binary image data obtained by the binarization processing is stored in the image memory 43. Store (S1).
FIG. 3 is a diagram illustrating an example of a binary image (a target of image processing by the image processing device Z) obtained in step S1.
In FIG. 3, an image of two character strings including a first character string “71C” and a second character string “07B” is an example of a binary image of the detection target character string.
Next, the image processing unit 44 extracts and labels the connected pixels included in the binary image based on the binary image data stored in the image memory 43 (S2).

[Step S3]
Next, the image processing unit 44 reads the binary image data from the image memory 43, and based on the binary image data, a predetermined character image condition (hereinafter referred to as “character image condition”) for each connected pixel in the binary image. , Character image condition) is satisfied, and among the connected pixels in the binary image, those that satisfy the character image condition are identified as first character candidate images (S3, first character candidate) Example of specific procedure). The identification result is stored in a memory (not shown) included in the image processing unit 44 in association with the label information of the connected pixels (identification information of the connected pixels).
For example, when it is guaranteed that the distance between the camera 3 and the image c of the detection target character string is almost constant, the character image condition is determined based on the size (height and width) of the circumscribed rectangle of the connected pixels. , A condition that the height FH and the width FW of the circumscribed rectangle of the character in the character image condition are within a predetermined error range, or the area (number of pixels) of the connected pixels is within a predetermined range. It is conceivable that the condition is
Further, when the distance between the camera 3 and the image c of the detection target character string can vary relatively greatly, the character image condition is determined based on the aspect ratio of the circumscribed rectangle of the connected pixels. The condition that the ratio is within a predetermined error range with respect to the ratio between the height FH and the width FW (aspect ratio) of the rectangle, or the area ratio between the connected pixel and its circumscribed rectangle, or the perimeter of the connected pixel It is conceivable that the ratio between the square of and the area is within a predetermined range.
It is also conceivable to specify connected pixels that simultaneously satisfy a plurality of the above-described conditions as the first character candidate image. Information on the character image condition is stored in advance in the general-purpose memory 46, and the image processing unit 44 refers to the information.
FIG. 4 is a diagram in which the circumscribed rectangles of the connected pixels ch11 to ch16 identified as the first character candidate images by the process of step S3 in the binary image shown in FIG. 3 are represented by broken lines.

In this manner, in step S3 (the first character candidate specifying procedure), the image processing unit 44 determines whether or not the character image condition is satisfied from all the connected pixels included in the binary image. By discriminating, those that are not likely to represent character images are excluded, and those that can be candidates for character images are specified (selected). The connected pixels specified in step S3 (the first character candidate images ch11 to ch16) may include noise images that do not actually represent character images. In FIG. 4, the image ch16 of the first character candidate corresponds to the noise image.
Further, among the plurality of character images, those having poor image quality (for example, those in the connected state or the broken state) are not specified as the first character candidate images in this step S3 (leak from the candidates). There is a case. In FIG. 4, the first character “7” in the first column of the detection target character string leaks from the candidate character image because it is in the broken state.

[Step S4]
Next, the image processing unit 44 determines that the relative positional relationship among the connected pixels (images of the first character candidates) identified in step S3 is relative to the character in the character arrangement condition (see FIG. 5). That satisfy a specific positional relationship within a predetermined error range is specified as a second character candidate image (S4, an example of the second character candidate specifying procedure). The identification result is stored in a memory (not shown) included in the image processing unit 44 in association with the label information of the connected pixels (identification information of the connected pixels).
More specifically, the image processing apparatus 4 sequentially selects one candidate from the first character candidate images as a leader ch1s, and each of the leader ch1s and all the other first character candidate images ch1i. And whether the relative positional relationship satisfies both of the condition expressed by the following formula (1) and the condition expressed by the formula (2) (hereinafter referred to as a character placement allowable condition). The determination result is stored in a memory included in the image processing unit 44.
Then, after all the first character candidate images are set as the leader ch1s and the discrimination process is completed, any of the first character candidate images among the first character candidate images is designated as the leader ch1s. Even in such a case, an image satisfying the character arrangement permission condition is always specified (selected) as the second character candidate image (corresponding to the character candidate image).
Alternatively, every time a leader ch1s is selected, all other first character candidate images in which the positional relationship between the leader ch1s and the leader ch1s satisfies the character placement permission conditions are set as one group, After setting the group with each of the first character candidate images as the leader ch1s, the group having the largest number of first character candidate images (number of elements in the group) belonging to the group is selected, and the group It is also conceivable to specify the image of the first character candidate belonging to as the image of the second character candidate.
FIG. 6 is a diagram showing the positional relationship between the leader ch1s and the image ch1i of the other first character candidate whose position is to be compared with the leader ch1s.
In step S4, the image processing unit 44 measures the dimension W (character width) of the circumscribed rectangle of the leader ch1s in the column direction and the dimension orthogonal to the column direction of the circumscribed rectangle of the leader ch1s (hereinafter referred to as the column orthogonal direction). A difference dW in the column direction position between H (character height) and the reference point P0s in the leader ch1s and the reference point P0i in the image ch1s of the other first character candidate to be compared with the leader; The position difference dH in the column orthogonal direction between the reference point P0s of the leader ch1s and the reference point P0i of the other image ch1s of the first character candidate is detected.
In the example shown in FIG. 6, the reference points P0s and P0i are one vertex in the circumscribed rectangle of the first character candidate image (connected pixel), but the same positions for all the first character candidate images. As long as it is a condition point, another point (for example, the center point of the circumscribed rectangle) may be the reference point.

（１）式及び（２）式において，変数ｍは前記文字配置条件における列方向の文字数ＮＣに対し−（ＮＣ−１）〜＋（ＮＣ−１）の範囲の整数であり，変数ｎは前記文字配置条件における列数ＮＬに対し−（ＮＬ−１）〜＋（ＮＬ−１）の範囲の整数である。また，同式において，ＦＷ，ｄＦＷ，ＦＨ，ｄＦＨは，図５に示した前記文字配置条件を表す既知の情報（寸法）である。また，同式において，定数ｄｎは前記文字配置条件における一列の文字数ＮＣに対応するマージン値（許容誤差）であり，定数ｄｍは前記文字配置条件における列数ＮＬに対応するマージン値（許容誤差）である。これらマージン値ｄｎ，ｄｍの情報は，予め前記汎用メモリに記憶され，前記画像処理部４４がその情報を参照する。なお，本実施形態における前記検出対象文字列の画像は，一列分の文字列における文字相互間の位置ずれ，及び一列分の文字列相互間の前記列直交方向における位置ずれはほとんど生じない一方，一列文の文字列相互間の列方向における位置ずれは大きくなり得る。このため，定数ｄｎに比べて定数ｄｍの方が大きな値が設定される。
この（１）式及び（２）式は，ステップＳ３で特定した前記第１文字候補の画像（連結画素）の中から，その相対的な位置関係が前記文字配置条件における文字相互の相対的な位置関係（前記検出対象文字列における文字相互の相対的な位置関係）を所定の誤差範囲内で満たすか否かを判別するための条件式の一例である。 Further, in step S4, the image processing unit 44 determines whether or not both of the following expressions (1) and (2) representing the character placement permission condition are satisfied, thereby determining the image of the second character candidate. Is identified.

In the equations (1) and (2), the variable m is an integer in the range of-(NC-1) to + (NC-1) with respect to the number of characters NC in the column direction in the character arrangement condition, and the variable n is It is an integer in the range of-(NL-1) to + (NL-1) with respect to the number of columns NL in the character arrangement condition. In the above equation, FW, dFW, FH, and dFH are known information (dimensions) representing the character arrangement conditions shown in FIG. In the above equation, the constant dn is a margin value (allowable error) corresponding to the number of characters NC in one column in the character arrangement condition, and the constant dm is a margin value (allowable error) corresponding to the number of columns NL in the character arrangement condition. It is. Information on these margin values dn and dm is stored in advance in the general-purpose memory, and the image processing unit 44 refers to the information. In the image of the detection target character string in the present embodiment, the positional deviation between characters in the character string for one column and the positional deviation in the column orthogonal direction between the character strings for one column hardly occur. The positional deviation in the column direction between character strings in a single column sentence can be large. For this reason, the constant dm is set to a larger value than the constant dn.
The expressions (1) and (2) are obtained by comparing the relative positional relationship between the characters in the character arrangement condition among the images (concatenated pixels) of the first character candidate specified in step S3. It is an example of a conditional expression for determining whether or not a positional relationship (relative positional relationship between characters in the detection target character string) is satisfied within a predetermined error range.

As described above, in step S4, the image processing unit 44 determines that the relative positional relationship among the images (concatenated pixels) of the first character candidate specified in step S3 is the expression (1) and ( 2) By determining whether or not the expression is satisfied, those that are not likely to represent character images are excluded, and those that can be character image candidates (images of the second character candidates) are specified ( select. As a result, a noise image at a position clearly distant from the position of the image of the detection target character string among the connected pixels can be excluded from the character image candidates. By the process of step S4, for example, in FIG. 4, the first character candidate images ch11 to ch15 satisfying the expressions (1) and (2) are specified as the second character candidate images. The image ch16 of the first character candidate that is an image is not specified as the image of the second character candidate.
As described above, in step S3 and step S4, both the condition of the connected pixel alone (the character image condition) and the arrangement condition between the plurality of connected pixels are discriminated, and the condition for specifying the character image candidate is strict. As a result, each of the images of the second character candidates is in a state that can be said to be an image of a character constituting the detection target character string almost certainly. However, the stricter the conditions for specifying a character image candidate, the more likely that connected pixels representing the character image will leak out of the character image candidate due to its image quality and position disturbance.

[Step S5]
Next, the image processing unit 44 determines the position (position of the reference point) and the size (W, H) of each of the second character candidate images (corresponding to the character candidate images) that are the connected pixels identified in step S4. And an area including all the images of the second character candidate in the binary image, based on the character arrangement condition and a predetermined variation range of the character arrangement, and including the entire image of the detection target character string (S5, an example of the character string inclusion region estimation procedure). The area estimated in step S5 is hereinafter referred to as a character string inclusion estimation area. Further, the estimation result (coordinate information representing the character string inclusion estimation area) by the process of step S5 is stored in a memory (not shown) provided in the image processing unit 44.
FIG. 7 is a diagram illustrating an example of an image of the character string inclusion estimation area A1 specified in step S5. For example, the image processing unit 44 estimates the character inclusion estimation area A1 according to the following procedure.
First, the image processing unit 44 calculates the position of the center point Q0 of the circumscribed rectangular area A0 of the entire image of the second character candidate.
Further, the image processing unit 44 calculates the average value of the interval dW in the column direction of each of the reference points P0i of the image of the second character candidate, the number of characters NC in one column in the character arrangement condition, and the variation range in the column direction. A value obtained by multiplying a predetermined coefficient (a coefficient greater than 1) is calculated as the length La1 in the column direction of the character string inclusion estimation area A1.
Further, the image processing unit 44 calculates an average value of the interval dH in the column orthogonal direction of each reference point P0i of the second character candidate image, the number of columns NL in the character arrangement condition, and the column orthogonal direction in the column orthogonal direction. A value obtained by multiplying by a predetermined coefficient (coefficient greater than 1) representing the fluctuation range is calculated as a column orthogonal width Wa1 of the character string inclusion estimation area A1.
Then, the image processing unit 44 sets the range of the length La1 in the column direction and the range of the width Wa1 in the column orthogonal direction centered on the center point Q0 as the character string inclusion estimation region A1.

As described above, all of the images of the second character candidates identified in step S4 are in a state where it can be said that all of them are character images that constitute the detection target character string almost certainly.
Therefore, if region estimation is performed based on the position and size of the image of the second character candidate, the character placement condition, and the predetermined variation range of the character placement, the images of all the characters constituting the detection target character string. Can be estimated (specified). The process of specifying the character string inclusion estimation area A1 is the process of step S5.
As shown in FIG. 7, the image of the character string inclusion estimation area A1 also includes an image of the character “7” that has not been specified as the character candidate image in steps S3 and S4. However, the character string inclusion estimation area A1 includes an extra area corresponding to the predetermined variation range of the character arrangement with respect to the actual existence area of the image of the detection target character string.
Thereafter, the image processing unit 44 executes a character image existing area specifying process for the image of the character string inclusion estimation area A1.

[Steps S6 and S7]
Next, the image processing unit 44 performs the processing of steps S6 and S7 described below (an example of a predetermined processing) on the image of the character string inclusion estimation area A1 in the binary image, thereby performing the detection. The existence range in the column orthogonal direction orthogonal to the column direction of each of the character string images constituting the target character string is estimated (S6, S7: an example of the column width range estimation procedure).
That is, the image processing unit 44 first calculates the distribution of projection data in the column direction for the image of the character string inclusion estimation area A1 in the binary image (S6, an example of the column direction projection data calculation procedure).
Next, the image processing unit 44, based on the calculation data in step S6 (distribution of projection data in the column direction), the image of the character string for one column that constitutes the detection target character string in each column orthogonal direction. An existence range (hereinafter referred to as a column range) is estimated (S7, an example of the range estimation procedure). Note that the calculation result (projection data associated with the position in the column direction) by the process in step S6 and the estimation result (coordinate information representing the column range) by the process in step S7 are not shown in the image processing unit 44. Stored in the memory.
FIG. 8 is a diagram illustrating an example of the image of the character string inclusion estimation area A1 and the distribution of the projection data Ix in the column direction in the image.
As can be seen from FIG. 8, in the range in the column orthogonal direction (character height range) where the image of the character string exists, the value of the projection data Ix in the column direction becomes high, and the range between the character string columns (gap) , The value of the projection data Ix in the column direction becomes low.
The image processing unit 44 identifies the position (position in the column orthogonal direction) that forms the valley in the distribution of projection data Ix in the column direction, and sets that position as the delimiter position ym for each character string image for one column. Thus, each column range is specified (estimated).
In the example shown in FIG. 8, in the distribution of the projection data Ix in the column direction, the center position of the range where the projection data falls below a predetermined threshold value Ixs (the range between y1 and y2) is defined as a plurality of the column ranges A11. , A12. The threshold value Ixs is, for example, a value obtained by multiplying the average value of the projection data Ix in the column direction by a predetermined coefficient (coefficient greater than 0 and smaller than 1), or a predetermined constant. .
Moreover, you may perform the process shown as a step which cuts out a character string in patent document 1 in this step S7.
In addition, when there is little need to consider the positional deviation in the column orthogonal direction of the image of the character string for one column, the character string inclusion estimation area A1 is equally divided by the number NL of known character strings. It is also conceivable to estimate the column ranges A11, A12 (existing ranges in the column orthogonal direction of the image of one column of character strings).

[Step S8]
Next, the image processing unit 44 calculates the distribution of projection data in the column orthogonal direction for each of the images in the column ranges A11 and A12 (hereinafter referred to as column region images g11 and g12) in the image of the character string inclusion region A1. (S8). The calculation result (projection data associated with the position in the column orthogonal direction) by the processing in step S8 is stored in a memory (not shown) provided in the image processing unit 44.
FIG. 9 is a diagram illustrating an example of the distribution of the column area images g11 and g12 in the character string inclusion estimation area A1 and the projection data Iy1 and Iy2 (hereinafter referred to as first projection data) in the column orthogonal direction in the image.
As can be seen from FIG. 9, the values of the first projection data Iy1 and Iy2 are high (in peaks) in the column direction range (character width range) where the character image exists for each of the column area images g11 and g12. , The values of the first projection data Iy1 and Iy2 are low (at valleys) in the range between the characters (gap). Further, since the width and interval of the character image in the column direction are substantially constant, the width and pitch of the peaks of the first projection data Iy1 and Iy2 are substantially constant in the range where the character string image exists.
On the other hand, when a noise image exists, the values of the first projection data Iy1 and Iy2 become high (in a mountain) also in the range where the noise image exists. However, normally, the pitch of the peaks of the first projection data Iy1 and Iy2 corresponding to the image differs between the character image and the noise image.

なお，（３）式において，添え字ｉは，前記列領域画像ｇ１１，ｇ１２の識別番号を表し，Ｉｙi(ｘ)は，列方向の位置ｘにおける前記第１投影データの値，Ｉｙ1’(ｘ)及びＩｙ2’(ｘ)は，列方向の位置ｘにおける前記補正第１投影データの値，Ｉｙimaxは，ｉ番目の前記列領域画像における前記第１投影データのピークレベルの指標値，αは予め定められた係数（０より大きく１未満の定数）である。ここで，ピークレベルの指標値Ｉｙimaxは，例えば，ｉ番目の前記列領域画像における前記第１投影データの最大値や，或いは同データにおける最も値の大きいものから所定個数分のデータの平均値などである。
図９には，（３）式に基づく補正処理後の前記補正第１投影データＩｙ1’，Ｉｙ2’の分布のグラフ（黒塗りグラフ）も示されている。
この補正により，文字画像に対応する前記補正第１投影データＩｙ1’，Ｉｙ2’が，文字画像それぞれの形状の違いに起因する値の変化が少ないデータとなり，文字画像が存在する範囲であるか否かを２値的に表すデータとなる。
本実施形態では，移行の処理において，列直交方向の投影データについて，その上限を制限する補正処理を施したデータを用いて文字画像領域の特定処理を行う。しかしながら，その補正を行わずに列直交方向の投影データそのものを用いて以降に示す文字画像領域の特定処理を実行することも考えられる。但し，上限を制限する補正処理を施したデータを用いた方が，文字画像領域をより正確に特定する上で好適である。 [Step S9]
Next, the image processing unit 44 performs a correction process for limiting the upper limit value for each of the first projection data Iy1 and Iy2 (projection data in the column orthogonal direction) calculated in step S8, and the corrected data (hereinafter, referred to as “correction data”). The corrected first projection data Iy1 ′ and Iy2 ′ are calculated (S9). The calculation result (corrected projection data associated with the position in the column orthogonal direction) by the processing in step S9 is stored in a memory (not shown) provided in the image processing unit 44. Steps S8 and S9 are an example of the first column orthogonal direction projection data calculation procedure.
For example, the image processing unit 44 calculates the corrected first projection data Iy1 ′, Iy2 ′ from the first projection data Iy1, Iy2 based on the following equation (3).

In equation (3), the subscript i represents the identification number of the row region images g11 and g12, Iyi (x) is the value of the first projection data at the position x in the row direction, Iy1 ′ (x ) And Iy2 ′ (x) are values of the corrected first projection data at the position x in the column direction, Iyimax is an index value of the peak level of the first projection data in the i-th column region image, and α is previously A predetermined coefficient (a constant greater than 0 and less than 1). Here, the peak level index value Iyimax is, for example, the maximum value of the first projection data in the i-th row area image, or the average value of a predetermined number of pieces of data from the largest value of the data. It is.
FIG. 9 also shows a distribution graph (black graph) of the corrected first projection data Iy1 ′ and Iy2 ′ after the correction processing based on the expression (3).
By this correction, the corrected first projection data Iy1 ′ and Iy2 ′ corresponding to the character image become data with little change in value due to the difference in the shape of each character image, and whether or not the character image exists in the range. This is binary data.
In the present embodiment, in the transition process, the character image region specifying process is performed using the data subjected to the correction process for limiting the upper limit of the projection data in the column orthogonal direction. However, it is also conceivable to execute the character image region specifying process described below using the projection data itself in the column orthogonal direction without performing the correction. However, it is preferable to use data that has been subjected to correction processing for limiting the upper limit in order to more accurately identify the character image area.

[Steps S10 and S11]
Next, the CPU 41 applies the corrected first projection data Iy1 ′ and Iy2 ′ calculated in step S9 (data subjected to correction processing for limiting the upper limit value for the first projection data calculated in step S8). Based on this, a cross-correlation function with the column-direction shift amount k as a variable is derived (S10, an example of the cross-correlation function deriving procedure). The column-direction shift amount k is synonymous with the column-direction position shift amount or the column-direction position offset amount.
In this step S10, the CPU 41 selects one reference row area image from among the plurality of row area images g11, g12, the corrected first projection data for the reference row image area, and the like. The cross-correlation function R (k) between the corrected first projection data for each of all the column image regions is derived.
FIG. 10 is a graph showing an example of the cross-correlation function R (k) of the corrected first projection data (data obtained by correcting the projection data in the column orthogonal direction) of the row area image.
As can be seen from FIG. 10, the cross-correlation function R (k) of the corrected first projection data has a peak (mountain) whose peak is the maximum correlation value Rmax with a certain shift amount kp, and almost the center of the peak. It has a value distribution in which a relatively low peak that occurs at a symmetrical position (a position with the shift amount k as a coordinate) is formed.
If the size of the character image and the interval between the character images in the character string for one column are substantially constant, the shift amount k is an image of the character string for one column (the column region image) in the cross-correlation function R (k). The correlation value R becomes the highest when it matches the mutual shift amount in the column direction. In addition, this phenomenon is not easily affected by disturbances in the image quality of some character images.
Therefore, in the next step S11, the CPU 41 detects (derived) a shift amount kp corresponding to the maximum correlation value Rmax for each of the cross-correlation functions R (k), and uses the shift amount kp as a reference row region image. Is set as a (mutual) column-direction positional shift amount Δxi (hereinafter referred to as a column mutual shift amount) between each of the other column region images (S9, an example of the column mutual shift amount derivation procedure). The setting result Δxi (= kp) is stored in the general-purpose memory 46.
Note that the column mutual shift amount for the reference column region image is zero.

[Step S12]
Next, the CPU 41 is based on the inter-peak correlation interval wk (see FIG. 10) of the correlation value in the cross-correlation function R (k) obtained by the process of step S10, and the number of characters NC in one row in the character arrangement condition. Then, the length L in the column direction (hereinafter referred to as column length) of the range where the image of the detection target character string exists in each of the column area images g11 and g12 is calculated (S12, of the column length calculation procedure). One case).
For example, the CPU 41 detects, for each of the plurality of cross-correlation functions R (k), an interval wk (hereinafter referred to as peak interval) between a peak having the maximum correlation value Rmax as a vertex and a peak adjacent thereto. An average value of the interval wk is calculated.
Further, the CPU 41 calculates a value obtained by multiplying the average value of the peak interval wk and the number of characters NC in one row, or a value obtained by multiplying the value by a predetermined margin rate β (a constant greater than 1), The column length L is calculated. Alternatively, a value obtained by multiplying the maximum value of the peak interval wk and the number of characters NC in one column may be calculated as the column length L.
If the size of the character image and the character image interval in one character string are substantially constant, the peak interval wk in the cross-correlation function R (k) corresponds to the pitch of the character image. Therefore, by multiplying the peak interval wk by the number of characters NC, the column length L (length in the column direction of the image of the character string for one column) can be calculated.

[Steps S13 and S14]
Next, the image processing unit 44 shifts (corrects the position) the corrected first projection data Iy1 ′ and Iy2 ′ calculated in step S9 by the mutual shift amount Δxi in the column direction (S13). Hereinafter, data obtained by shifting the corrected first projection data Iy1 ′ and Iy2 ′ by Δxi is referred to as post-shift corrected first projection data Iy1 ″ and Iy2 ″.
Further, the image processing unit 44 integrates the post-shift corrected first projection data Iy1 ″, Iy2 ″ in the column orthogonal direction (integration for each position in the column direction), and the image processing unit 44 includes the data after the integration ( (S14).
The post-integration data calculated in step S14 has an upper limit value for projection data in the column orthogonal direction in the entire image obtained by correcting (shifting) the position of each of the plurality of column region images g11 and g12 by the column shift amount Δxi. This corresponds to the distribution of data subjected to the limiting correction process. The data after integration calculated in step S14 is hereinafter referred to as corrected second projection data Iyt ′. Note that step S14 is an example of the second column orthogonal direction projection data calculation procedure.
FIG. 11 shows data g21 and g22 in which the positional deviation in the column direction of the column region images g11 and g12 is corrected (shifted by Δxi) and data obtained by correcting the upper limit of projection data in the column orthogonal direction of the entire image (the correction second It is a figure showing an example of distribution of projection data Iyt ').
As shown in FIG. 11, in the entire image g2 after misalignment correction composed of the images g21 and g22 after misalignment correction in the column direction, the arrangement positions of the character images in each column are substantially the same in the column orthogonal direction. It becomes an image. Further, the position where the peak of the value is formed in the corrected second projection data Iyt ′ coincides with the position of the character image in the entire image g2 after the positional deviation correction.
In this way, in the detection target character string, if the character size and the character spacing in the character string for one column are substantially constant, the column region images g21 and g22 after the correction of the positional deviation in the column direction are In any case, the position of the gap between the character image and the character image (character delimiter position) is almost the same. Accordingly, the plurality of column region images g21 and g22 whose column direction misalignment is corrected are based on the projection data in the column orthogonal direction in the plurality of column region images g2 whose column direction misalignment is corrected. It is possible to specify the position where character images are separated.

[Step S15]
Next, the image processing unit 44 calculates the moving average of the column length L based on the reference position while changing the reference position in the column direction with respect to the calculation data (the corrected second projection data) obtained in step S14. A moving average value for the section is calculated (S15). That is, the image processing unit 44 calculates an average value within the column length L with respect to each of the plurality of reference positions in the column direction for the corrected second projection data.
For example, the image processing unit 44 uses the left end position (one end position in the column direction) of the row area images g11 and g12 as the start point of the reference position, and sequentially shifts the reference position to the right (for example, one pixel at a time). Every time (shift), the average value (that is, the moving average value) of the corrected second projection data in the section of the column length L is calculated from the reference position in the right direction (column direction).
Here, unless the noise image is so large as to exceed the character image, normally, when all the character images are included in the average interval of the moving average value, the average value of the data in the interval (the moving average value) And the integrated value is maximized.
Therefore, the image processing unit 44 uses the column region images g11, g12 based on the reference position and the column length L when the moving average value (average value of the section having the column length L) is maximized. A processing target range (a range to be specified as a character image existing area) is specified in each of them (S15, an example of the row area image processing target range specifying procedure).
For example, the image processing unit 44 specifies the range of the column length L from the reference position xk when the moving average value is maximized in the right direction (column direction) as the processing target range Wx (see FIG. 11). ). The processing results of steps S14 and S15 are stored in a memory (not shown) provided in the image processing unit 44.
The processing target range Wx specified in step S15 is a range obtained by removing an extra region in the column direction (a region where no character image exists) from the character string inclusion estimation region A1 estimated in step S5.
It should be noted that “average value” in steps S14 and S15 may be replaced with “integrated value”, and the result is the same even if such replacement is performed.

[Step S16]
Next, the image processing unit 44, based on the corrected second projection data Iyt 'calculated in the process of step S14, the presence of each character image constituting the detection target character string in each of the column area images g11 and g12. The position of the area is specified (S16, an example of the character area specifying procedure). The identification result is stored in a memory (not shown) provided in the image processing unit 44. At that time, the image processing unit 44 specifies the existence area of each character image constituting the detection target character string with respect to the processing target range Wx specified by the process of step S15.
As can be seen from FIG. 11, in the range in the column direction (character width range) where the character string image exists, the value of the corrected second projection data Iyt ′ increases, and the character image and the character image (gap) In this range, the value of the corrected second projection data Iyt ′ becomes low. Therefore, in step S16, the image processing unit 44 specifies the position (column direction position) of the portion forming the valley in the corrected second projection data Iyt ', and determines the position as the delimiter position of each character image in each column. Thus, the existence area of each character image is specified. The image processing unit 44 specifies the range in the column orthogonal direction of each of the column image regions g11 and g12 (the range estimated in step S7) as the range in the column orthogonal direction of each character image.
In FIG. 11, each area indicated by reference numeral A3 represents an example of the existence area of each character image specified in step S16. As shown in FIG. 11, the existence area A3 of each character image specified in step S16 includes an area in which an image of the character “7” not specified as the character candidate image in steps S3 and S4 exists. . As described above, according to the present invention, it is possible to correctly specify the existence area of a character image having a disordered image.
As described above, if the process of specifying the position of the character delimiter in the column direction is performed based on the projection data in the column orthogonal direction for the entire plurality of column region images g21 and g22 whose positional deviation is corrected, the detection is performed. When some character images in the image c of the target character string are disturbed, the influence of the disturbance of some character images compared to the process of specifying the character image area individually for each of the row area images g11 and g12. As a result, the character image delimiter position can be specified more correctly.

[Steps S17 and S18]
Next, the image processing unit 44 executes a process (character image extraction process) of cutting out a character image from each of the character image existence areas A3 specified in step S16 (S17). The extracted (extracted) character image is stored in the image memory 43.
Next, the CPU 41 executes a pattern recognition process with a predetermined character candidate image for each of the character images cut out in step S17, so that the cut-out character image represents any character. A character recognition process is performed to identify whether or not (S18).
In step S18, for example, the image processing unit 44 causes the neural network learned using the predetermined character candidate image data as teacher data to input the character image data extracted in step S17. Thus, the character recognition process is executed. It is assumed that information defining the neural network is stored in advance in the general-purpose memory 46.
In addition, the character recognition processing may be considered in which character candidate image data is stored in advance in the general-purpose memory 46, and pattern matching processing is performed between each of the image data and the character image data cut out in step S17.

[Step S19]
Finally, the CPU 41 executes an output process of the result of the character recognition process in step S19 (S19). For example, the CPU 41 transmits (outputs) the result of character recognition processing to the host computer through the communication unit 47, displays (outputs) the result on the display unit 45, or records (outputs) it in the general-purpose memory 46. ) Is executed.
According to the above processing (character image image processing method), when specifying the area of each character image for the image of the detection target character string made up of a plurality of character strings, the image of each character string for one column When the mutual displacement in the column direction is large, or when the gap between the character image is not clear for some of the characters constituting the detection target character string (for example, 2 as shown in FIG. 12 Even in the case of a value image, the existence area of each character image can be specified as accurately as possible.

    The present invention can be used for image processing of character images.

The block diagram showing the schematic structure of the image processing apparatus Z of the character image which concerns on embodiment of this invention. 5 is a flowchart showing a procedure of character image cutout / character recognition processing by the image processing apparatus Z. FIG. 3 is a diagram illustrating an example of a binary image that is an object of image processing by the image processing apparatus Z. The figure showing an example of the candidate of the character image in a binary image. The figure which represented typically the character arrangement conditions in a detection object character string. The figure showing the positional relationship between character image candidates in a binary image. The figure showing an example of the image of the character string inclusion estimation area specified by the image processing device Z. The figure showing an example of the image of a character string inclusion estimation area | region, and the projection data distribution of the column direction in the image. The figure showing an example of the line area image in a character string inclusion estimation area, and the projection data distribution of the line orthogonal direction in the image. The graph showing an example of the cross correlation function of the data which correct | amended the projection data of the row orthogonal direction of the row | line | column area | region image. The figure showing an example of distribution of the data which correct | amended the upper limit of the projection data of the row orthogonal direction of the image which corrected the position shift of the row direction of a row | line | column area | region image, and the whole image. The figure showing an example of a binary image with inferior image quality about a part of character image.

Explanation of symbols

Z: Image processing device for character image c: Image of character string to be detected 1: Managed member 2: Conveying means 3: Camera 41: CPU
42: Camera interface 43: Image memory 44: Image processing unit 45: Display unit 46: General-purpose memory 47: Communication unit A1: Character string inclusion estimation area A11, A12: Column range g1, g2: Column area image

Claims (7)

An image obtained by imaging a member composed of a plurality of character strings, on which an image of a character string to be detected, in which the character size, the number of characters in one row, the character interval, the number of columns, and the column interval satisfy a predetermined character arrangement condition, is formed. The binary image data of the character string is read out from a predetermined storage means, and based on the read data, the existence area of each character image constituting the detection target character string in the binary image is specified, and An image processing method for a character string image that extracts a character image from an existing area of the image,
With a given processor
A first character candidate specifying procedure for specifying, based on the read data, a connected pixel in the binary image that satisfies a predetermined character image condition;
Among the connected pixels specified by the first character candidate specifying procedure, specify those whose relative positional relationship satisfies the relative positional relationship between characters in the character arrangement condition within a predetermined error range. A second character candidate identification procedure;
In the binary image, based on the position and size of each of the character candidate images that are the connected pixels specified by the second character candidate specifying procedure, the character arrangement condition, and a predetermined variation range of the character arrangement. A character string inclusion region estimation procedure for estimating a region including all of the character candidate images and including the entire image of the detection target character string;
By performing a predetermined process on the image of the character string inclusion estimation region that is the region estimated by the character string inclusion region estimation procedure in the binary image, a character string for one column that constitutes the detection target character string Column width range estimation procedure for estimating the existence range in the column orthogonal direction orthogonal to the column direction of each of the images of
A correction process for limiting the upper limit value of the projection data in the column orthogonal direction or the projection data in the column orthogonal direction is performed on the column area image that is an image of each range estimated by the column width range estimation procedure in the image of the character string inclusion region A first column orthogonal direction projection data calculation procedure for calculating a distribution of data;
A cross-correlation function derivation procedure for deriving a cross-correlation function having a shift amount in the column direction as a variable, based on calculation data obtained by the first column orthogonal direction projection data calculation procedure;
A column cross-shift amount derivation procedure for deriving a column cross-shift amount that is a position shift amount in the column direction between the column region images based on the cross-correlation function obtained by the cross-correlation function derivation procedure;
Calculating the distribution of the projection data in the column orthogonal direction in the entire image obtained by correcting the position of each of the plurality of column region images by the column mutual shift amount or the distribution of the data subjected to the correction process for limiting the upper limit value of the projection data; Column orthogonal direction projection data calculation procedure,
A character region specifying procedure for specifying a position of an existing region of each character image constituting the detection target character string in each of the column region images, based on calculation data obtained by the second column orthogonal direction projection data calculating procedure;
A character string image processing method characterized by automatically executing The column width range estimation procedure includes:
A column direction projection data calculation procedure for calculating a distribution of projection data in the column direction for the image of the character string inclusion estimation region in the binary image;
A range estimation procedure for estimating the existence range in the column orthogonal direction of each of the character string images constituting the detection target character string based on the calculation data by the column direction projection data calculation procedure;
The image processing method for a character string image according to claim 1, comprising: The processor is
Based on the interval between peaks of a plurality of correlation values in the cross-correlation function obtained by the cross-correlation function derivation procedure and the number of characters in one row in the character arrangement condition, the detection target character string in each of the row region images A column length calculation procedure for calculating the length in the column direction of the range where the image exists,
For the calculation data by the second column orthogonal direction projection data calculation procedure, the integrated value of the values within the range of the column length, which is the calculation result by the column length calculation procedure with respect to each of a plurality of reference positions in the column direction, or A column region image processing target range specifying procedure for automatically calculating a mean value and specifying a processing target range in the column region image based on the reference position and the column length when the calculated value is maximized. Run
2. The character region specifying procedure specifies an existence region of an image of each character constituting the detection target character string for the processing target range specified by the column region image processing target range specifying procedure. The image processing method of the character string image in any one of 2 or 2. In the column cross-deviation amount guiding out procedure will derive the column cross-deviation amount on the basis of the shift amount in the column direction corresponding to the maximum correlation value in said cross-correlation function deriving procedure the cross-correlation function obtained by The image processing method of the character string image in any one of Claims 1-3.   The predetermined character image conditions are the size of the circumscribed rectangle of the connected pixel, the aspect ratio of the circumscribed rectangle of the connected pixel, the area of the connected pixel, the area ratio of the connected pixel and its circumscribed rectangle, and the surrounding of the connected pixel. The image processing method for a character string image according to any one of claims 1 to 4, wherein at least one of the ratio of the square of the length and the area thereof is within a predetermined range. An image obtained by imaging a member composed of a plurality of character strings, on which an image of a character string to be detected, in which the character size, the number of characters in one row, the character interval, the number of columns, and the column interval satisfy a predetermined character arrangement condition, is formed. The binary image data of the character string is read out from a predetermined storage means, and based on the read data, the existence area of each character image constituting the detection target character string in the binary image is specified, and A character string image image processing program for causing a predetermined processor to execute a procedure for extracting a character image from an existing area of an image,
A first character candidate specifying procedure for specifying, based on the read data, a connected pixel in the binary image that satisfies a predetermined character image condition;
Among the connected pixels specified by the first character candidate specifying procedure, specify those whose relative positional relationship satisfies the relative positional relationship between characters in the character arrangement condition within a predetermined error range. A second character candidate identification procedure;
In the binary image, based on the position and size of each of the character candidate images that are the connected pixels specified by the second character candidate specifying procedure, the character arrangement condition, and a predetermined variation range of the character arrangement. A character string inclusion region estimation procedure for estimating a region including all of the character candidate images and including the entire image of the detection target character string;
By performing a predetermined process on the image of the character string inclusion estimation region that is the region estimated by the character string inclusion region estimation procedure in the binary image, a character string for one column that constitutes the detection target character string Column width range estimation procedure for estimating the existence range in the column orthogonal direction orthogonal to the column direction of each of the images of
A correction process for limiting the upper limit value of the projection data in the column orthogonal direction or the projection data in the column orthogonal direction is performed on the column area image that is an image of each range estimated by the column width range estimation procedure in the image of the character string inclusion region A first column orthogonal direction projection data calculation procedure for calculating a distribution of data;
A cross-correlation function derivation procedure for deriving a cross-correlation function having a shift amount in the column direction as a variable, based on calculation data obtained by the first column orthogonal direction projection data calculation procedure;
Based on the cross-correlation function obtained by the cross-correlation function derivation procedure, a column cross-shift amount calculation procedure for calculating a column cross-shift amount that is a position shift amount in the column direction between the column region images;
Calculating the distribution of the projection data in the column orthogonal direction in the entire image obtained by correcting the position of each of the plurality of column region images by the column mutual shift amount or the distribution of the data subjected to the correction process for limiting the upper limit value of the projection data; Column orthogonal direction projection data calculation procedure,
A character region specifying procedure for specifying a position of an existing region of each character image constituting the detection target character string in each of the column region images, based on calculation data obtained by the second column orthogonal direction projection data calculating procedure;
A character string image image processing program for causing a predetermined processor to execute. An image obtained by imaging a member composed of a plurality of character strings, on which an image of a character string to be detected, in which the character size, the number of characters in one row, the character interval, the number of columns, and the column interval satisfy a predetermined character arrangement condition, is formed. The binary image data of the character string is read out from a predetermined storage means, and based on the read data, the existence area of each character image constituting the detection target character string in the binary image is specified, and An image processing device for a character string image that extracts a character image from an existing area of the image,
First character candidate specifying means for specifying, based on the read data, a connected pixel in the binary image that satisfies a predetermined character image condition;
Among the connected pixels specified by the first character candidate specifying means, specify those whose relative positional relationship satisfies the relative positional relationship between characters in the character arrangement condition within a predetermined error range. A second character candidate specifying means;
In the binary image, based on the position and size of each of the character candidate images that are the connected pixels specified by the second character candidate specifying means, the character arrangement conditions, and a predetermined variation range of the character arrangement. A character string inclusion region estimation means for estimating a region including all of the character candidate images and including the entire image of the detection target character string;
By performing a predetermined process on the image of the character string inclusion estimation area, which is an area estimated by the character string inclusion area estimation means in the binary image, a character string for one column constituting the detection target character string Column width range estimation means for estimating the existence range in the column orthogonal direction orthogonal to the column direction of each of the images of
A correction process is performed to limit the upper limit value of projection data in the column orthogonal direction or projection data in the column orthogonal direction with respect to the row area image that is an image of each range estimated by the row width range estimation unit in the image of the character string inclusion region. First column orthogonal direction projection data calculation means for calculating data distribution;
Cross-correlation function deriving means for deriving a cross-correlation function having the shift amount in the column direction as a variable based on the calculation data by the first column orthogonal direction projection data calculating means;
A column cross-shift amount calculating unit that calculates a column cross-shift amount that is a position shift amount in the column direction between the column region images based on the cross-correlation function obtained by the cross-correlation function deriving unit;
Calculating the distribution of the projection data in the column orthogonal direction in the entire image obtained by correcting the position of each of the plurality of column region images by the column mutual shift amount or the distribution of the data subjected to the correction process for limiting the upper limit value of the projection data; Column orthogonal direction projection data calculation means,
A character area specifying means for specifying a position of an existing area of each character image constituting the detection target character string in each of the row area images based on calculation data by the second column orthogonal direction projection data calculating means;
An image processing apparatus for character string images, comprising:

Images