JP4162195B2 - Image processing apparatus and image processing program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
  The present invention relates to a character recognition process used in an OCR (optical character reader) or the like, and more specifically, a process of extracting character candidates to be character recognition based on a character string image read from a document original. In the extraction of the character line used in the previous stage, the line to be excluded from one recognition target group, such as the ruby line for the body line, is detected, and the character line to be used is output using the detection result. Have means that allow toImage processingapparatusAnd image processingRegarding the program.
[0002]
[Prior art]
In a conventional OCR (optical character reading device), processing for recognizing characters written on a document based on an image read from a document document by a scanner is performed. When this processing is performed, character units are cut out in order to extract character candidates that are targets for character recognition based on the character string image included in the read image. The process of cutting out character lines from the image is performed in the previous stage. Since this line cut-out defines the recognition target, proper cut-out is necessary to guarantee the recognition accuracy.
In cutting out character lines, a conventionally known method is a method that should be called a “circumscribed rectangle integration method”. In this method, first, a connected pattern of black pixels that are grouped as a graphic is extracted from an input character string image, and a circumscribed rectangle is obtained for each extracted pattern, and then these rectangles are converted into one line. Integration is performed in accordance with an integration rule for determining the constituent elements (for example, integration is performed if the horizontal and vertical distances between rectangles are within a predetermined range), and rows are cut out using the obtained row rectangles (Japanese Patent No. 2895122, reference).
[0003]
When a line is cut out by this “surrounding rectangle integration method”, if a line corresponding to a note such as ruby is added to a normal character line constituting the text in the target document, the ruby line has been used so far. Etc. were usually cut out in the same way as normal lines.
By the way, in recent character recognition devices, there is a case where a character recognition result obtained by the pattern matching method is corrected by some language processing and a recognition result having a plausible form as a document is finally output. Many. In performing such language processing, if a ruby line is cut out in the same way as a normal line, there will be no linguistic connection before and after the ruby line, so the accuracy of correction by language processing is greatly reduced.
For example, when an image with a ruby line added as shown in FIG. 9 is input, if the ruby line is cut out in the same way as a normal line, the language processing indicates that “Today is a fine day, tomorrow is a cloudy day. Since the sentence “Nari” is targeted, correct language processing cannot be performed, leading to a reduction in recognition performance of the character recognition device.
[0004]
[Problems to be solved by the invention]
  Therefore, in order to prevent ruby from being mixed into the text, Japanese Patent Laid-Open No. 8-101886 (character recognition device) proposes a method for removing ruby lines. In the method disclosed in Japanese Patent Application Laid-Open No. 8-101886, the ruby character line to be removed is added between the lines, and therefore the final line is the text line, and the final line is used as a reference. The ruby lines are detected in reverse order from the last line to the previous character line and the first line. However, this precondition does not always hold, and the final line is not always a normal line. For example, in a manuscript with footnotes and the like, there are cases in which characters having the same size as ruby are arranged on the last line. Therefore, Japanese Patent Laid-Open No. 8-101886 detects these lines corresponding to the case where lines such as ruby and footnote that are to be excluded from one recognition target group (text line group) or to be handled separately are in the last line. Can not do it.
  As described above, the conventional technique has a problem that it does not sufficiently meet the demand for improving the character string recognition (cutting) accuracy of a manuscript including ruby or footnote. The present invention,UpThe purpose of the present invention was made in view of the above-mentioned problems of the prior art, and its purpose is not to treat it as one recognition target group (text line group) like a ruby line or a footnote line with respect to the text line, or When detecting a line to be handled separately from one recognition target group (text line group), no matter what character line on the document the target line is (a precondition as described in JP-A-8-101886 is set). So that it can be detected unconditionally)DoMake it possibleImage processingapparatusAnd image processingTo provide a program.
[0005]
[Means for Solving the Problems]
  The invention of claim 1An image processing apparatus that outputs a character line from a character string image, wherein means for selecting a reference line from a plurality of character lines and whether each character line is ruby based on a shape value of the selected reference line Determining means, wherein the selecting means selects a reference line according to an evaluation value by a membership function having a line width and a line height as variables.
[0006]
  The invention of claim 2 is described in claim 1Image processingIn the deviceThe means for selecting selects a reference row according to a sum of an evaluation value by a membership function with a line width as a variable and an evaluation value by a membership function with a line height as a variable.It is characterized by doing.
[0008]
  The invention of claim 4A step of selecting a reference line from a plurality of character lines included in image data according to an evaluation value by a membership function having a line width and a line height as variables, and a shape of the selected reference line And a step of determining whether each character line is ruby based on a value..
[0009]
  The invention of claim 55. The image processing program according to claim 4, wherein the selecting step is performed according to a sum of an evaluation value by a membership function having a line width as a variable and an evaluation value by a membership function having a line height as a variable. It is characterized by selecting a reference line.
[0022]
DETAILED DESCRIPTION OF THE INVENTION
The character line data output means for improving the accuracy of character string recognition (cutting out), which is a constituent requirement of the present invention, includes a text body included in character lines included in a plurality of character string images input as processing targets. A line that you do not want to treat as a single recognition target group (text line group), such as a ruby line or a footnote line, or a line that you want to treat separately from a single recognition target group (text line group) Ruby line ”), and it is possible to distinguish and output as character line data for each of the text line and the ruby line, and provide means (procedure) for that purpose.
In each embodiment of the present invention described below, all the character lines included in the character string image of a plurality of lines are cut out, and the body line and the ruby line can be detected from among them. The detection procedure extracts a reference line (a line that can be regarded as a standard text line) from all the extracted lines according to a predetermined rule, and each extracted line is based on the shape value of the extracted reference line. This is based on the procedure of determining whether the line is a line belonging to the body line or not (that is, ruby line) and outputting the extracted line data using the determination result.
FIG. 1 is a block diagram showing a configuration of a processing apparatus (system) used in common for the implementation of each embodiment.
Referring to FIG. 1, 1 is an image input unit for reading a document image such as a scanner and inputting the image, and 3 is a character line for cutting out all the character lines included in the input character string image of a plurality of lines. The cutout unit 5 is a character line determination unit that determines whether or not each of the cut out lines is a line belonging to one recognition target group (text line group), and 7 is a determination result. A line output unit that outputs the extracted character line data.
In the following embodiments, modes related to the implementation will be exemplified with a focus on procedures relating to the output of character line data characterizing the present invention. Therefore, the block diagram shown in FIG. 1 also shows only the part related to line segmentation in the character recognition device, omits other components, and finally performs the entire character recognition process, that is, the input of the target image. The entire procedure (means) for obtaining the optimal solution for character recognition is not explained in the following, but the procedure known from the past as the basic procedure necessary for character recognition is applied to the entire character recognition process. The implementation is possible.
[0023]
“Embodiment 1”
The present embodiment relates to character line data output (detection) processing executed by the processing system shown in FIG. In the character line data output processing shown here, the maximum line width is used as a rule for selecting a reference line (a line that can be regarded as a standard character line), and the shape value of the selected / extracted reference line is high. Based on the above, it is determined whether each extracted line is a line belonging to the body line or a ruby line, its attribute, and for the line determined to be a ruby line, the line data is deleted and the extracted line is deleted. An embodiment of a processing process according to a procedure of outputting data will be described.
FIG. 2 shows a flowchart of the character line data output process of this embodiment.
Referring to FIG. 2, in the flow of the present embodiment, first, the image input unit 1 inputs a plurality of character string images to be recognized into the character line cutout unit 3 (S11). In this input, the recognition target area data may be given together with the image. When a recognition target region is given, only the inside of the given region may be set as a row cutout target.
Next, the character line cutout unit 3 cuts out a character line by applying a conventionally proposed method (S12). Various methods such as a method using projection have been proposed for extracting character lines. Here, the “surrounding rectangle integration method” shown in the above-mentioned section “Prior Art” is used. For example, when line segmentation is performed by the method shown in Japanese Patent No. 2895122, the coordinates of the line rectangle obtained by integration and the coordinates of the rectangle in the line (the circumscribed rectangle of the black pixel connected component that is the basis of integration) are the character line segmentation unit. 3 and sent to the character line determination unit 5. Note that all the row data relating to the row cut out in this step is stored in the storage unit 2.
[0024]
Next, the character line determination unit 5 that receives the line cutout result determines whether each cut out line is a line belonging to the body line or not (that is, whether it is a ruby line). As this procedure, first, one reference line is selected from all the cutout lines sent from the character line cutout unit 3, and this is determined as a criterion for determination. In selecting the reference row, the row having the widest width among all the row rectangles is set as the reference row using the coordinates of the row rectangle (S13). According to the method of defining this reference line, the width of ruby lines usually does not become wider than the width of the text line to which ruby is assigned, so if this criterion is selected, lines that can be regarded as standard text lines are displayed. Selected and no ruby line is selected. After setting the reference line, as a procedure for setting the reference value used for determination, the height value: Sh of the line having the maximum line width determined as the reference line is acquired, and half of the acquired line height value: Sh / 2 is set as a reference value for determination (S14).
Next, in the determination of each cut-out line, a line whose height: H is lower than Sh / 2 is determined as a ruby line, and the other line is determined as a body line. In the present embodiment, a process of deleting line data determined to be a ruby line is performed. Therefore, as a procedure of this process, it is determined whether or not the height H of each row is lower than Sh / 2 (S15). If it is lower than Sh / 2 (S15-YES) The ruby row data is deleted from the row data previously stored in the storage unit 2 (S16). In this ruby line determination / line data deletion process, determination is made for all the lines for each cut-out line, and therefore the processes of steps S15 and S16 are repeatedly executed for the number of lines.
After applying the ruby line determination / line data deletion process to each cut-out line, the ruby line data is deleted, and line data including information on the line rectangle and the in-line rectangle corresponding to the other body lines is passed through the line output unit 7. The data is output to a subsequent processing unit for performing character recognition processing (S17), and this processing is terminated.
[0025]
“Embodiment 2”
The present embodiment relates to character line data output (detection) processing executed by the processing system shown in FIG. The character line data output process shown here introduces a membership function with the line width and line height as variables as a rule for selecting the reference line (a line that can be regarded as a standard character line). The evaluation value as is calculated.
FIG. 5 is a diagram showing an example of the membership function, and FIG. 4 is a diagram for explaining the shape value of the cut row rectangle used as a parameter for setting the membership function.
This membership function has the following conditions (1) and (2):
(1) The wider the line width, the higher the evaluation value.
(2) The lower the line height, the higher the evaluation value.
Set according to.
Here, as a function satisfying the above (1), as shown in the example of FIG. 5A, a linear function with an evaluation value of maximum row width: MaxW as a maximum value: 1 is used.
Further, as a function satisfying the above (2), as shown in the example of FIG. 5 (B), a linear function is used in which the evaluation value of the maximum row height: MaxH is the minimum value: 0. However, when the row height is smaller than a predetermined threshold value: Thignore, so as not to select a line-only line or a noise-only minute line by mistake, or to prevent a ruby line from being selected as a reference line. The evaluation value is set to “0”. In consideration of the continuity of the membership function, the evaluation value is set to the maximum value: 1 at a height MaxH / 2 that is half the maximum row height MaxH.
Using this membership function, the evaluation value of each target row is calculated. Here, the evaluation value is calculated using the sum of the evaluation value obtained from the row height membership function and the evaluation value obtained from the row width membership function as the evaluation value for each row, and the row with the largest evaluation value as the reference row. (Refer to the description of the processing flow shown in FIG. 3 described later).
Also, based on the height as the shape value of the selected / extracted reference line, it is determined whether each extracted line is a line belonging to the body line, a ruby line, or its attribute, and is determined as a ruby line. The character line data output processing process is executed according to the procedure of deleting the line data and outputting the extracted line data for the line that has been cut.
[0026]
FIG. 3 shows a flowchart of the character line data output process of this embodiment.
Referring to FIG. 3, in the flow of the present embodiment, first, the image input unit 1 inputs a plurality of character string images to be recognized into the character line cutout unit 3 (S 21). In this input, the recognition target area data may be given together with the image. When a recognition target region is given, only the inside of the given region may be set as a row cutout target.
Next, the character line cutout unit 3 cuts out the character line (S22). The character line segmentation method is implemented by applying the “circumscribed rectangle integration method” in the same manner as described in “Embodiment 1”. The coordinates of the line rectangle obtained as a result of cutting out the character line and the coordinates of the rectangle in the line (the circumscribed rectangle of the black pixel connected component that is the basis of integration) are output from the character line cutting unit 3 and sent to the character line determination unit 5. Sent. Note that all the row data relating to the row cut out in this step is stored in the storage unit 2.
Next, the character line determination unit 5 that receives the line cutout result determines whether each cut out line is a line belonging to the body line or not (that is, whether it is a ruby line). As this procedure, first, one reference line is selected from all the cutout lines sent from the character line cutout unit 3, and this is determined as a criterion for determination.
In selecting a reference row, the above-described membership function is applied to obtain an evaluation value, and the row having the maximum evaluation value is selected as the reference row.
[0027]
FIG. 6 is a diagram for explaining the reference row selection processing. (A) in the figure shows a plurality of rows S1 to S5 to be subjected to recognition processing, and (B) and (C) in the figure are members set by the method described above (see FIGS. 4 and 5). The operation state when the membership function is applied to the ship function and the target row shown in FIG.
As a procedure for selecting a reference row, first, a membership function is set (S23). For this purpose, the maximum row width MaxW and the maximum row height MaxH are extracted from the plurality of rows S1 to S5 that are the targets of recognition processing (see FIG. 6A). A membership function for the line width (see FIG. 6B) is set using the extracted maximum line width MaxW as a parameter, and a membership function for the line height (FIG. 6C) using the extracted maximum line height MaxH as a parameter. Set the reference).
Thereafter, an evaluation value: membership value V of each target row is calculated using the set membership function, and a row having the maximum value Vmax is selected as a reference row. Therefore, first, Vmax = 0 is set, and initial conditions in this process are set (S24).
Next, a membership value V is calculated by applying a membership function to each of the target rows S1 to S5 (S25). As shown in the examples of FIGS. 6B and 6C, membership values according to the function are obtained for the row width value and the row height value of each of the target rows S1 to S5. However, here, the sum of the membership values for each of the row width value and the row height value is calculated and set as the finally obtained membership value V.
Further, since the row having the maximum value Vmax is selected, the membership value V sequentially obtained for each row is compared with the maximum value Vmax of the rows obtained so far (S26), and according to the result, that is, the maximum value Vmax. Is changed (S26-YES), the row data of the maximum value Vmax after the change (the height value of the row having the maximum row width used in the subsequent stage: Sh) is updated (S27). In this reference line selection process, determination is made for all the lines S1 to S5 for each cut-out line, and therefore the processes in steps S25 to S27 are repeatedly executed for the number of lines.
[0028]
After the reference line is determined by the reference line selection process, it is determined whether each extracted line is a line belonging to the body line (ie, whether it is a ruby line). As the reference value used for the determination, the row height value Sh as the row data of the reference row acquired in the previous step S27 is used, and half of this row height value: Sh / 2 is used as the determination reference value. Set.
In the determination of each cut-out line, a line whose height: H is lower than Sh / 2 is determined as a ruby line, and the other line is determined as a body line. In the present embodiment, a process of deleting line data determined to be a ruby line is performed. Accordingly, as a procedure of this processing, it is determined whether or not the height H of each row is lower than Sh / 2 (S28). If the height H is lower than Sh / 2 (S28-YES). The ruby row data is deleted from the row data previously stored in the storage unit 2 (S29). In this ruby line determination / line data deletion process, determination is made for all the lines for each cut-out line, and therefore the processes in steps S28 and S29 are repeatedly executed for the number of lines.
After applying the ruby line determination / line data deletion process to each cut-out line, the ruby line data is deleted, and line data including information on the line rectangle and the in-line rectangle corresponding to the other body lines is passed through the line output unit 7. The data is output to a subsequent processing unit for performing character recognition processing (S30), and this processing is terminated.
[0029]
“Embodiment 3”
The present embodiment relates to character line data output (detection) processing executed by the processing system shown in FIG. The character line data output process shown here relates to the improvement of the above-described “Embodiment 2”. The improvement is that it is possible to suppress over-detection of the ruby line, and the height of the line rectangle similar to the ruby line (H <Sh / 2 in the case of the above embodiments). It is more appropriate to consider it as a body line, not a ruby line, among those belonging to a line having a height), that is, there is a case where an adverse effect may occur if it is deleted as a ruby line. A process is added so that a line that is over-detected in can be treated as a body line. As a means for this, even if the ruby line is determined by the line height check, the distance between the previous and next lines is wider than the reference line height, that is, the ruby line cannot be clearly determined (note that In the case of original ruby lines and noise lines, the interval between the previous and next lines is often very narrow, so adding this condition does not affect the detection of ruby lines) It is assumed that processing means for excluding those that are treated as ruby lines is used.
In addition, as a rule for selecting a reference line (a line that can be regarded as a standard character line), a membership function with the line width and line height as variables is introduced, and the evaluation value as the reference line is calculated by this function. Then, it is the same as “Embodiment 2”.
[0030]
FIG. 7 shows a flowchart of the character line data output process of this embodiment.
Referring to FIG. 7, in the flow of the present embodiment, steps S31 to S37 are performed until a reference row is selected by evaluation using a membership function, and a row height value Sh possessed by the reference row is set as a reference value for ruby row determination. The processing procedure is performed in the same manner as the procedure of the “embodiment 2” described above (steps S21 to S27 in FIG. 3). Accordingly, the description of the processing procedure of steps S21 to S27 of the above-described “Embodiment 2” is referred to, and description of this processing procedure is omitted here.
Whether or not each extracted line is a line belonging to the body line after the reference line is determined by the reference line selection process (S35 to 37) using the line having the maximum evaluation value by the membership function as the reference line Or determine its attributes. In the present embodiment, this determination is performed in two stages, i.e., ruby line determination based on line height and determination based on an interval between before and after (or up and down) lines performed to correct over-detection of ruby lines.
Here, the reference value used for the ruby line determination based on the line height uses the line height value Sh as the line data possessed by the reference line acquired in the previous step S37, and is half of this line height value. : Sh / 2 is set as a reference value for determination, and a line whose height: H is lower than Sh / 2 is determined as a ruby line. In addition, the determination based on the distance between the previous and next lines is determined by comparing the distance between the previous line and the previous line (the distance between the previous line and the next line) compared to the height Sh of the reference line: Think of it as a body line.
Based on the result of the determination of each cut-out line in two stages, processing is performed to output line data of a text line or a line that is regarded as a text line, and to delete line data determined to be other ruby lines.
[0031]
Therefore, as a procedure in this processing flow, first, an interval (interval with the previous row + interval with the next row) B between each row is calculated (S38).
Next, it is determined whether or not the height H of each row is lower than half the height of the reference row Sh / 2 (S39). If the row is lower than Sh / 2 (S28-YES). Further, it is determined whether or not the interval B between the previous and subsequent rows calculated in step S38 is wider than the height Sh of the reference row (S40).
Here, when the interval B between the preceding and succeeding rows is narrower than the height Sh of the reference row (S40-YES), since it is determined that there is no over-detection, the data of this row is stored in the storage unit 2 first. The deleted row data is deleted (S41). In this ruby line determination / line data deletion process, determination is made for all the lines for each cut-out line, and therefore the processes in steps S38 to S41 are repeatedly executed for the number of lines.
After applying the ruby line determination / line data deletion process to each cut-out line, the ruby line data determined to be a ruby line with no over-detection is deleted, and the line rectangle of the other lines that are considered as body lines or body lines Then, the line data including the information of the in-line rectangle is output to the subsequent processing unit for performing the character recognition process through the line output unit 7 (S42), and this process is terminated.
[0032]
“Embodiment 4”
The present embodiment relates to character line data output (detection) processing executed by the processing system shown in FIG. The character line data output process shown here is a modification of “Embodiment 3” described above. The point to be modified is that in “Embodiment 3”, over-detection is suppressed, an obvious ruby line is determined, and line data is deleted for the determined ruby line. Without deleting the data, an output process is performed so that the data can be used in a subsequent character recognition process as data of a different system from the body line.
In this ruby line output process, information indicating that it is a ruby line is added, and the line data is output to the subsequent process through the line output unit 7. In the subsequent processing, it is possible to perform post-processing such as language processing by ignoring the ruby line by the information indicating that the ruby line is added. In addition, separately from the processing, each ruby line can be processed independently to obtain a recognition result, and finally combined with the recognition result of the main text line and output as a processing result of the character recognition device. For output, a method suitable for use, such as outputting a recognition result including a ruby portion in a format corresponding to ruby, such as RTF, may be adopted.
[0033]
FIG. 8 shows a flowchart of the character line data output process of this embodiment.
Referring to FIG. 8, in the flow of this embodiment, a reference row is selected by evaluation using a membership function, a row height value Sh possessed by the reference row is set as a reference value for ruby row determination, and The processing procedure of steps S51 to S60 until the interval B is obtained and overdetection is suppressed and an obvious ruby line is determined is the same as the above-described procedure of “Embodiment 3” (steps S31 to S40 in FIG. 7). It carries out similarly. Accordingly, the description of the processing procedure of steps S31 to S40 of the above-described “Embodiment 3” is referred to, and description of this processing procedure is omitted here.
It is determined whether or not the interval B between the preceding and succeeding rows is narrower than the height Sh of the reference row for the rows determined as ruby rows through the processing up to step S59 (S60). Is smaller than the reference line height Sh, it is determined that the ruby line is clear (no over-detection) (S60-YES). Here, with respect to the cut-out line determined to be an obvious ruby line, the determination result that the line is a clear ruby line without deleting the line data as in the above-described “embodiment 3”. It is added to the line rectangle of the line (including information on the rectangle within line) (S61).
After applying ruby line determination / line data addition processing to each cut-out line, information on the determination result is added to ruby lines determined to be obvious ruby lines, and text lines or text lines other than obvious ruby lines For the line regarded as a line, the line data including the original line rectangle information and the in-line rectangle information is output to the subsequent processing unit for performing the character recognition processing through the line output unit 7 (S62), and this processing is terminated. To do.
[0034]
“Embodiment 5”
This embodiment shows other embodiment of the character recognition apparatus concerning this invention.
An apparatus configured using a general-purpose computer is exemplified as means for executing the processing including the character line data output processing procedure shown in the above-described “Embodiment 1” to “Embodiment 4”.
Since it is implemented by a general-purpose computer, as an element, an input unit I / F for an input device such as a scanner, a keyboard, and a mouse, an auxiliary storage device such as a CPU, a storage device, and a hard disk drive, and an output device for a display It is equipped with components that a normal computer has such as an output I / F, a removable storage medium drive, a removable storage medium, a controller for communicating with other devices via a network, and these are connected by a bus (device ( System).
In addition, the above-mentioned “Embodiment 1” to “Implementation” for realizing the character string recognition (cutting out) function according to the present invention on a part of the storage medium used by the storage device and the auxiliary storage device such as a hard disk drive A program (software) for executing each processing procedure shown in the character recognition method including the character line data output processing procedure shown in “Mode 4” is recorded.
The character string image to be processed is input by reading a document with an input device such as a scanner, and is stored in, for example, a hard disk. The CPU reads a program that realizes the processing procedure described above from a recording medium included in the storage unit, executes processing according to the program on the target character string image, and outputs the processing result and the like to the display.
Note that the character recognition apparatus according to the present invention may be implemented in such a manner that a part of the functions is provided on the network by connecting to an external apparatus via a network by a network controller.
[0035]
【The invention's effect】
  The reference line can be selected more accurately from multiple character lines.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a character string recognition (cutout) processing system according to the present invention.
FIG. 2 is a flowchart of character line data output processing according to “Embodiment 1”;
FIG. 3 is a flowchart of character line data output processing according to “Embodiment 2”;
FIG. 4 is a diagram illustrating parameters for setting a membership function for obtaining a reference row.
FIG. 5 is a diagram showing an example of a membership function set using the parameters shown in FIG.
FIG. 6 is a diagram for explaining reference row selection processing based on membership values;
FIG. 7 shows a flowchart of character line data output processing according to “Embodiment 3”.
FIG. 8 is a flowchart of character line data output processing according to “Embodiment 4”;
FIG. 9 shows an example of an image to which a ruby line is added.
[Explanation of symbols]
1 ... image input unit, 2 ... storage unit,
3 ... character line segmentation unit, 5 ... character line determination unit,
7: Line output unit.

Claims (4)

An image processing apparatus that outputs a character line from a character string image,
Means for selecting a reference line from a plurality of character lines;
Means for determining whether each character line is ruby based on the shape value of the selected reference line,
The image processing apparatus according to claim 1, wherein the selecting means selects a reference row according to an evaluation value based on a membership function having a row width and a row height as variables.   The selection means selects a reference row according to a sum of an evaluation value by a membership function having a line width as a variable and an evaluation value by a membership function having a line height as a variable. The image processing apparatus according to 1. On the computer,
Selecting a reference line from a plurality of character lines included in the image data according to an evaluation value by a membership function using a line width and a line height as variables;
Determining whether each character line is ruby based on the shape value of the selected reference line;
An image processing program for executing   The selection step includes selecting a reference row according to a sum of an evaluation value by a membership function having a line width as a variable and an evaluation value by a membership function having a line height as a variable. 3. The image processing program described in 3.

Images