JP3998439B2 - Image processing apparatus, image processing method, and program causing computer to execute these methods - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image processing apparatus, an image processing method, and a program for causing a computer to execute these methods when an input image is compared with a reference image to collate both images, and noise components included in the input image are removed. In particular, the present invention relates to an image processing apparatus, an image processing method, and a program for causing a computer to execute these methods to efficiently and accurately determine a form when the form is determined using characters in the form.
[0002]
[Prior art]
Conventionally, a technique for discriminating a form without discriminating a discrimination code or a discrimination mark when discriminating the type of the form is known. For example, Japanese Patent Application Laid-Open No. 4-268865 discloses a ruled line from input form image data. A form that extracts horizontal and vertical line segments and divides them into multiple areas, and uses the direction, length, and position of the line segments extracted for each area as a vector pattern for comparison with the standard pattern feature vector. A class type discrimination method is disclosed.
[0003]
If a line segment is extracted and used as a feature value as in this prior art, the line segment is interrupted due to scanner characteristics, rotation correction, etc., so it is necessary to perform interpolation processing to connect the line segments. Doing so may lead to the connection of separate line segments.
[0004]
For this reason, there has been proposed a technique for accurately discriminating forms by preventing a decrease in discrimination accuracy due to image fluctuations. For example, Japanese Patent Application No. 2000-95514 filed by the present applicant includes There is disclosed a form discriminating apparatus that discriminates a form type by using the ratio of black pixels located in pixel rows adjacent to each other in the horizontal and vertical directions.
[0005]
However, when there are a plurality of forms having the same ruled line feature, it is impossible to distinguish them. In this prior art, the form discrimination is performed by recognizing the characters included in the image.
[0006]
[Problems to be solved by the invention]
However, as in this prior art, when character recognition is performed by simply cutting out characters from an image, there is a problem that it is not possible to accurately identify a form when a lot of noise is included in the image. . In particular, there may be cases where the forms are copied using carbon paper, so there are many cases where characters cannot be read accurately due to rubbing of the carbon paper.
[0007]
For this reason, in a method of discriminating a form using characters in the form, how to accurately perform character recognition is an important issue in order to discriminate the form with high accuracy. In particular, since the card companies of the same series often use the same form with the ruled line form, the accuracy of character recognition greatly affects the discrimination accuracy of the form.
[0008]
The present invention has been made to solve the above-described problems caused by the prior art, and is an image that can efficiently and accurately determine a form when the form is determined using characters in the form. It is an object of the present invention to provide a processing apparatus, an image processing method, and a program for causing a computer to execute these methods.
[0009]
[Means for Solving the Problems]
In order to solve the above-described problems and achieve the object, the image processing apparatus according to the first aspect of the present invention includes a noise component included in the input image when the input image is compared with the reference image and both images are collated. In the image processing apparatus that removes the pixel, each pixel that forms a binary image obtained by binarizing the reference image and a pixel that forms a binary image of the input image corresponding to the pixel are both black pixels The pixel value of , Other pixel values are white pixels Template generating means for generating a template, Die-cutting is performed in which the pixel values of pixels other than the portion corresponding to each black pixel of the template generated by the template generation unit in the input image are white pixels. And a noise cut image generating means for generating a noise cut image obtained by removing a noise component from the input image.
[0010]
An image processing apparatus according to a second aspect of the present invention is the image processing apparatus according to the first aspect, wherein the template generation means applies a minimum value filter to the reference image to expand a black pixel portion; And binarizing means for binarizing an image to which the minimum value filter is applied by the minimum value filtering means.
[0011]
According to a third aspect of the present invention, there is provided an image processing apparatus according to the first or second aspect of the present invention, wherein among the pixels forming the noise cut image generated by the noise cut image generating means, the pixels cut out from the input image Each pixel having no value has a value obtained from the pixel values of a plurality of pixels in the input image that are not to be punched as a pixel value.
[0012]
According to a fourth aspect of the present invention, there is provided an image processing apparatus according to the first, second or third aspect of the present invention, wherein the input image is generated by the binarized black pixel number and the noise cut image generating means. The image processing apparatus further includes fraud determination means for determining whether the input image is an improper image based on a difference between the noise-cut image and the number of black pixels in the binary image.
[0013]
According to a fifth aspect of the present invention, there is provided an image processing method for removing a noise component included in the input image when comparing both images by comparing the input image with a reference image. Each pixel forming a binary image obtained by binarizing a pixel and a pixel forming a binary image of an input image corresponding to the pixel are both black pixels. , Other pixel values are white pixels A template generation process for generating a template, Die-cutting is performed in which the pixel values of pixels other than the portion corresponding to each black pixel of the template generated by the template generation unit in the input image are white pixels. And a noise cut image generation step of generating a noise cut image obtained by removing noise components from the input image.
[0014]
An image processing method according to a sixth aspect of the present invention is the image processing method according to the fifth aspect of the present invention, wherein the template generation step includes a minimum value filtering step of expanding a black pixel portion by applying a minimum value filter to the reference image; And a binarization step of binarizing the image to which the minimum value filter is applied by the minimum value filtering step.
[0015]
According to a seventh aspect of the present invention, there is provided an image processing method according to the fifth or sixth aspect, wherein pixels out of an input image among pixels forming a noise cut image generated by the noise cut image generation step are used. Each pixel having no value has a value obtained from the pixel values of a plurality of pixels in the input image that are not to be punched as a pixel value.
[0016]
According to an eighth aspect of the present invention, there is provided an image processing method according to the fifth, sixth or seventh aspect, wherein the input image is generated by the number of black pixels of a binary image obtained by binarizing the input image and the noise cut image generating step. The fraud determination step of determining whether or not the input image is a fraudulent image based on the difference between the noise cut image and the number of black pixels of the binary image is further included.
[0017]
According to a ninth aspect of the present invention, a program causes a computer to execute the method according to any one of the fifth to eighth aspects.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
Exemplary embodiments of an image processing apparatus, an image processing method, and a program for causing a computer to execute the method according to the present invention will be described below in detail with reference to the accompanying drawings. In the present embodiment, the case where the present invention is applied to a form classification device is shown.
[0019]
FIG. 1 is a functional block diagram showing the configuration of the form discriminating apparatus used in the first embodiment. The form discriminating apparatus 10 shown in the figure uses a character in the form in the second stage when the first stage form judgment using the feature amount of the ruled line and the accurate form discrimination cannot be performed in the first stage. This is a form discriminating apparatus for discriminating a form, and the feature of the present invention is particularly in the second stage form discrimination.
[0020]
In this first-stage form discrimination, if the feature amount of the reference image is registered in advance as a dictionary and the image of the form to be discriminated is input, the feature amount of the ruled line of this input image is extracted and By comparing, the type of form is determined.
[0021]
Here, the feature quantity used in the form discriminating apparatus 10 is a black pixel ratio in consideration of the ruled lines forming the essential contents of the form preprinted on the form. In this apparatus 10, line segment interpolation processing and the like are performed. Not done. The reason for this is that if such line segment interpolation processing is performed, the discrimination accuracy may be lowered. The black pixel ratio is a ratio of black pixels included in a pixel column in a predetermined section in the horizontal direction or vertical direction from the target pixel, and is a value obtained for each pixel of the image data.
[0022]
In the second stage form discrimination, the operator inputs characters included in the specific area of the form as text together with the reference image, and the characters included in the specific area of the input image are compared with the text characters. To make a detailed discrimination. The reason for performing such detailed determination is that, for example, the form type may not be determined only by the ratio of black pixels considering the ruled lines, such as the ruled lines being exactly the same and only the printed characters being different.
[0023]
Specifically, the form discriminating apparatus 10 applies a minimum value filter to the reference character string image to expand the black pixel portion of the reference character string image, binarizes it, and inputs the binary image and the input image. Create a template with a character string image. Then, by cutting out the input character string image using this template, a noise cut image in which noise components are removed from the input character string image is generated. Then, after recognizing the character in the noise cut image, it is determined whether or not the character matches by comparing it with a text character string that has been previously inputted and stored by the operator.
[0024]
As shown in FIG. 1, the form discriminating apparatus 10 includes an image input unit 101, a ruled line feature extracting unit 102, a dictionary creating unit 103, a ruled line feature dictionary 104, a specific area dictionary 105, and a ruled line feature matching unit 106. A detailed determination unit 107 and an output unit 108.
[0025]
The image input unit 101 is a scanner that optically inputs form image data. When the image data input unit 101 is set to a “reference image input mode” in which a reference image is input by a switching button (not shown) or the like, the input density image data and a pixel value with white pixels of “0” are set. Thus, binary image data in which the black pixel has a pixel value of “1” is output to the ruled line feature extraction unit 102. On the other hand, when the “collation mode” for collating the form is set, the input density image data is output to the detail determination unit 107 and the binary image is output to the ruled line feature extraction unit 102. .
[0026]
The ruled line feature extraction unit 102 is a processing unit that extracts ruled line features (features) from the binary image data received from the image input unit 101. Specifically, in the reference image input mode, the ruled line feature and density image are extracted. Data is output to the dictionary creation unit 103. In the collation mode, the ruled line feature is output to the ruled line feature collating unit 106.
[0027]
When the dictionary creation unit 103 receives the ruled line feature and density image data of the reference image from the ruled line feature extraction unit 102, the dictionary creation unit 103 creates or adds the ruled line feature dictionary 104 and the specific area dictionary 105 based on the information. It is.
[0028]
Specifically, when the dictionary creation unit 103 creates or adds the ruled line feature dictionary 104, the ruled line feature and the form type are associated with each other and registered in the ruled line feature dictionary 104.
[0029]
On the other hand, when the dictionary creation unit 103 creates or adds the specific area dictionary 105, the black pixel portion is expanded by applying a minimum value filter to the image data (reference character string image) of the specific area of the form. Then, the binary image data binarized and the text character string previously input by the operator are registered in the specific area dictionary 105 in association with the form type.
[0030]
The ruled line feature dictionary 104 is a dictionary that stores ruled line features in association with each form type, and the specific area dictionary 105 is a binary image of a specific area and text characters input by the operator for each form type. A dictionary in which columns are stored in association with each other.
[0031]
The ruled line feature matching unit 106 checks the ruled line feature (feature amount) of the image data of the document to be discriminated against the ruled line feature (feature value) of each reference image stored in the ruled line feature dictionary 104, and determines the image to be discriminated. A processing unit that selects a plurality of candidates based on the distance to the data and outputs the selected candidates to the detail determination unit 107.
[0032]
Note that a technique widely used in conventional character recognition or the like can be applied as such collation processing, and for example, identification can be performed based on the Euclidean distance or the like.
[0033]
The detail determination unit 107 is a processing unit that determines in detail which one of the plurality of candidates received from the ruled line feature matching unit 106 is closest to the input image, and is a form to be determined received from the image input unit 101. The image data of the specific area (input character string image) is cut out from the image data, and the text character string obtained by character recognition of the input character string image is collated with the text character string registered in the specific area dictionary 105.
[0034]
For example, as shown in FIGS. 7A and 7B, when the ruled lines of two forms are exactly the same, the type of the form cannot be determined only by the ruled line characteristics, and therefore the input image and the reference image Characters (character strings, logos, etc. that form the form such as form title and company name) included in each specific area are cut out and compared.
[0035]
However, the detailed determination unit 107 is not only to compare and collate the input character string image and the text character string of the reference image, but also to resist noise in consideration of the case where noise is included in the input character string image. Judgment is made. A detailed description of this point will be described later.
[0036]
The output unit 108 is a processing unit that outputs the determination result received from the detail determination unit 107. As the determination result, a registered form closest to the form to be determined can be output, but a plurality of candidates can also be output in order.
[0037]
Next, ruled line feature extraction processing by the ruled line feature extraction unit 102 shown in FIG. 1 will be described in more detail. FIG. 2 is an explanatory diagram for explaining the concept of ruled line feature extraction processing by the ruled line feature extraction unit 102 shown in FIG.
[0038]
As shown in FIGS. 9A and 9B, the ruled line feature extraction unit 102 has a section Pi (i = 1, 2, 3,..., K) (in the horizontal and vertical directions around the target pixel) ( The ratio of black pixels (black pixel ratio) included in the section length pi × 2 + 1 (dot)) is calculated.
[0039]
Specifically, in the case of the horizontal section 1 shown in FIG. 5A, the pixel values from the target pixel to the left and right pixels are examined. here,
Section length = 8 x 2 + 1 = 17 dots
Number of black pixels in the section = 11 dots
It becomes.
[0040]
However, in order to eliminate the influence of noise and vertical ruled lines (ruled lines in a direction different from the counting direction), the number of continuous black pixels is not counted. For example, in FIG. 5A, black pixels A and B are not counted because their continuous number is 1.
[0041]
For this reason,
Black pixel ratio = (11-2) / 17 = 0.529
It becomes. It should be noted that the horizontal directions 2 and 3 and the vertical direction shown in FIG.
[0042]
Thereafter, the form image is divided into M × N blocks, and the black pixel ratio of each pixel in the block is added to obtain a ruled line feature. Note that the number of dimensions of the ruled line feature is M × N × 2 (horizontal / vertical) × K dimensions.
[0043]
At this time, if the black pixel ratio is added only when it is larger than a certain threshold value, it is possible to eliminate fluctuation factors such as noise and handwritten characters. This is because handwritten characters and noise are a collection of short line segments compared to ruled lines, and the black pixel ratio in the section is also small.
[0044]
Next, an example of ruled line feature extraction by the ruled line feature extraction unit 102 shown in FIG. 1 will be described more specifically. FIG. 3 is an explanatory diagram illustrating an example of ruled line feature extraction by the ruled line feature extraction unit 102 illustrated in FIG. 1.
[0045]
In the case where there is a ruled line input image in the shape of “R” shown in FIG. 4A, if the section length is 3 dots and the threshold of the continuous number is not considered, each pixel in the horizontal direction is considered. The black pixel ratio is as shown in FIG. 4B, and the black pixel ratio of each pixel in the vertical direction is as shown in FIG.
[0046]
Then, when the image is divided into 3 × 3 blocks as shown in FIG. 4D and the black pixel ratio of each pixel in the horizontal direction shown in FIG. The ruled line feature shown in FIG. Further, when the black pixel ratio of each pixel in the vertical direction shown in FIG. 5C is added for each block, the ruled line feature shown in FIG.
[0047]
As described above, since the ruled line feature extraction unit 102 uses the black pixel ratio and the ruled line feature as the feature amount, it does not require a process for interpolating the break of the ruled line segment, and the ruled line feature extraction process 102 performs a process such as rotation correction. Even if the line segment is interrupted, the feature amount can be acquired stably.
[0048]
Further, if there are a plurality of sections as shown in FIG. 2, the features of ruled lines of various lengths can be obtained faithfully. Although not performed in the present embodiment, it is possible to suppress fluctuation due to rotation by performing a process of thickening the ruled line on the input image before feature extraction. In addition, it is possible to acquire a feature amount that is resistant to misalignment by applying various processes for increasing the recognition rate widely known in character recognition such as blurring.
[0049]
Next, a processing procedure for registering a form as a comparison target at the time of discrimination will be described. FIG. 4 is a flowchart showing a processing procedure when a form is registered as a dictionary for comparison at the time of discrimination.
[0050]
As shown in the figure, when registering a form as a comparison target at the time of discrimination, a form image is first taken from the image input unit 101 (step S401), and image preprocessing is performed as necessary. (Step S402). However, this pre-processing does not include line segment interpolation processing or the like.
[0051]
Thereafter, the ruled line feature extraction unit 102 calculates the black pixel ratio in the horizontal and vertical directions for the section designated in advance (step S403), and adds the black pixel ratio for each block to extract the ruled line feature (step S403). S404)
[0052]
Then, the dictionary creation unit 103 registers the ruled line feature extracted by the ruled line feature extraction unit 102 in the ruled line feature dictionary 104 (step S405), and then stores the ruled line feature in the ruled line feature dictionary 104 in the past. It is confirmed whether it can be determined by collation (steps S406 to S407).
[0053]
As a result, if it cannot be determined (No at step S407), the process of additionally registering the specific area information (binary image data and text character string of the specific area) in the specific area dictionary 105 is repeated (step S408). At this point (Yes in step S407), the process is terminated.
[0054]
For example, in the case of making a detailed determination using a character string, a character string (text data) in a specific area (character string such as a title or company name) with a characteristic on each form and its position are registered in advance. .
[0055]
By performing the above-described series of processing, the ruled line features of various forms are registered in the ruled line feature dictionary 104 and the binary image data and text data of the specific area are registered in the specific area dictionary 105 prior to the discrimination of the form. Can do.
[0056]
Next, a procedure for discriminating a form by the form discriminating apparatus 10 shown in FIG. 1 will be described. FIG. 5 is a flowchart showing a form discrimination processing procedure performed by the form discrimination apparatus 10 shown in FIG.
[0057]
As shown in the figure, when determining the type of a form, first, an image of the form is fetched from the image input unit 101 (step S501), and image preprocessing is performed as necessary (step S502). However, this pre-processing does not include line segment interpolation processing or the like.
[0058]
After that, the ruled line feature extraction unit 102 calculates the black pixel ratio in the horizontal and vertical directions for the section designated in advance (step S503), and adds the black pixel ratio for each block to extract the ruled line feature (step S503). S504)
[0059]
The ruled line feature matching unit 106 matches the ruled line feature extracted by the ruled line feature extracting unit 102 with the ruled line feature registered in the ruled line feature dictionary 104 (step S505), and the distance value is within a predetermined threshold value. It is checked whether or not there is, and the candidates for the form are sorted in the order of closeness according to this distance order.
[0060]
If it is within the predetermined threshold value, the detailed determination unit 107 performs detailed determination described later (step S506) and outputs the determination result (step S507). Outputs the determination result via the detailed determination unit 107 as it is (step S507).
[0061]
That is, among the candidates for the form, if the distance between the first place and the second place is more than a certain threshold, the first place is output as the determination result. It recognizes the character string of the area, and if it is still not good, it recognizes another specific area. The detailed determination processing procedure will be described later.
[0062]
By performing the above-described series of processing, it is possible to determine a form using the ruled line feature based on the ruled line feature dictionary 104 and the specific area dictionary 105 and the characters in the specific area.
[0063]
Next, the configuration of the dictionary creation unit 103 shown in FIG. 1 will be specifically described. FIG. 8 is a functional block diagram showing the configuration of the dictionary creation unit 103 shown in FIG. As shown in the figure, the dictionary creation unit 103 includes a ruled line feature dictionary creation unit 800 that creates a ruled line feature dictionary 104 and a specific area dictionary creation unit 810 that creates a specific area dictionary 105.
[0064]
The specific area dictionary creation unit 810 includes a cutout processing unit 811, a minimum value filtering unit 812, and a binarization processing unit 813. The minimum value filtering unit 812 corresponds to the minimum value filtering unit of claim 2, and the binarization processing unit 813 corresponds to the binarization unit of claim 2.
[0065]
The cutout processing unit 811 is a processing unit that cuts out a reference character string image located at a specific position in the reference image. Note that a specific position to be cut out is set in advance.
[0066]
The minimum value filtering unit 812 is a processing unit that applies a minimum value filter to the reference character string image and expands a black pixel portion in the reference character string image. Here, the minimum value filter is a mask pattern process in which a pixel having a minimum pixel value among neighboring pixels located in the vicinity of 8 of the target pixel is used as a pixel value of the target pixel.
[0067]
The binarization processing unit 813 is a processing unit that binarizes the density image to which the minimum value filter is applied by the minimum value filtering unit 812. In this binarization, a discrimination criterion method using a density histogram can be applied.
[0068]
Next, the processing procedure of the specific area dictionary creation unit 800 shown in FIG. 8 will be specifically described. FIG. 9 is a flowchart illustrating a processing procedure of the specific area dictionary creation unit 800 illustrated in FIG. 8, and FIG. 10 is an explanatory diagram illustrating an example of a character string image.
[0069]
As shown in FIG. 9, in the specific area dictionary creation unit 800, when a reference image is first input (step S901), the extraction processing unit 811 specifies a reference image as shown in FIG. 10A, for example. A reference character string image located in the area is cut out (step S902).
[0070]
Thereafter, the minimum value filtering processing unit 811 applies a minimum value filter to the reference character string image to expand the black pixel portion of the reference character string image as shown in FIG. 10B (step S903).
[0071]
Then, the binarization processing unit 813 binarizes the reference character string image to which the minimum value filter is applied using a discrimination criterion method or the like, and acquires a binary image as shown in FIG. 10C (step S904). . The binary image obtained in this way and a text character string separately input from the operator are stored in the specific area dictionary 105.
[0072]
Next, the configuration of the detail determination unit 107 shown in FIG. 1 will be specifically described. FIG. 11 is a functional block diagram illustrating a configuration of the detail determination unit 107 illustrated in FIG. As shown in the figure, the detail determination unit 107 includes a template creation unit 1101, a noise cut image extraction unit 1102, a fraud determination unit 1103, a character recognition processing unit 1104, and a character determination unit 1105. The template creation unit 1101 corresponds to the template generation unit of claim 1, and the noise cut image extraction unit 1102 corresponds to the noise cut image generation unit. The fraud determination unit 1103 corresponds to the fraud determination unit according to the fourth aspect.
[0073]
The template creating unit 1101 forms each pixel forming a binary image (binary image obtained by binarizing a reference character string image to which a minimum value filter is applied) read from the specific area dictionary 105, and an input character corresponding to the pixel A pixel that forms a binary image of a column image is a processing unit that creates a template in which the pixel values of pixels that are both black pixels are black pixels.
[0074]
The noise cut image extraction unit 1102 is a processing unit that extracts a pixel value of a template portion from an input character string image and extracts a noise cut image from which noise components are removed from the input character string image.
[0075]
The fraud determination unit 1103 is a processing unit that determines the possibility of fraud that exploits die cutting using a template according to the present invention. That is, when performing die cutting using a template, if all the pixels of the input character image are black pixels, each character of the reference character image is cut out due to the effect of the template. Therefore, the fraud determination unit 1103 regards such input images as being fraudulent and rejects them.
[0076]
The character recognition processing unit 1104 is a processing unit that recognizes and cuts out each character included in the noise cut image, and the character determination unit 1105 reads the character of the cut out noise cut image and the text character read from the specific area dictionary. It is a processing unit that compares the columns and determines whether or not each character matches.
[0077]
Next, the processing procedure of the detail determination unit 107 shown in FIG. 11 will be described. FIG. 12 is a flowchart illustrating a processing procedure of the detail determination unit 107 illustrated in FIG. 11, and FIG. 13 is an explanatory diagram illustrating an example of a character string image.
[0078]
As shown in FIG. 12, when an input image is input to the detail determination unit 107 (step S1201), the template creation unit 1101 cuts out an input character string image located in a specific area of the input image (step S1202). Specifically, unlike the reference image, the input image is often accompanied by a positional shift or the like. Therefore, a character string search area is provided slightly wider as shown in FIG. 13A, and thereafter, as shown in FIG. The input character string image is cut out. Note that a receipt stamp is added as noise to the input character string image.
[0079]
Thereafter, the template creation unit 1101 binarizes the input character string image using a discrimination criterion method or the like as shown in FIG. 13C (step S1203), and binarizes the input character string image. A template is created from the image and the binary image of the reference character string image extracted from the specific area dictionary 105 (step S1204).
[0080]
Specifically, each pixel that forms a binary image obtained by binarizing the reference character string image and a pixel that forms a binary image of the input character string image corresponding to the pixel are both black pixels. A template having a pixel value of black as a black pixel is generated.
[0081]
Thereafter, the noise cut image extraction unit 1103 cuts out the pixel value (density data) of the input character string image using this template and creates a noise cut image as shown in FIG. 13D (step S1205). ). Note that a representative density value of the high density class (white) when the density histogram of the input character string image is divided into two by the discrimination reference method is assigned to each pixel forming the background portion of the noise cut image as a pixel value.
[0082]
Further, the fraud determination unit 1103 binarizes the noise-cut image using a discrimination criterion method or the like as shown in FIG. 13E (step S1206), and how much noise component is removed. Check. Specifically, the number of black pixels of the binary image shown in FIG. 13E is subtracted from the number of black pixels of the binary image shown in FIG. 13C to obtain a change amount (step S1207). A value obtained by dividing the amount by the total number of pixels of the input character string image is obtained (step S1208).
[0083]
Then, it is checked whether or not this value is equal to or smaller than a predetermined value (step S1209). If it is not equal to or smaller than the predetermined value (No at step S1209), the input image is rejected (step S1210). In other words, if the value obtained by dividing the amount of change by the total number of pixels in the input character string image is large, the possibility of unauthorized use of template punching increases, so this value exceeds the specified value. Will be rejected as possibly fraudulent. Here, a value of about “0.07 (7%)” can be used as the predetermined value.
[0084]
Here, for convenience of explanation, explanation of character recognition and character collation performed thereafter is omitted, but these are performed using a known character recognition technique or the like.
[0085]
As described above, according to the present embodiment, the template creation unit 1101 corresponds to each pixel that forms a binary image obtained by binarizing the reference character string image to which the minimum value filter is applied, and the pixel. A template in which the pixel values of the pixels that form the binary image of the input character string image are both black pixels is created as a black pixel, and the noise cut image extraction unit 1102 extracts the template portion from the input character string image. Since the pixel value is cut out and the noise cut image from which the noise component is removed is extracted from the input character string image, the noise in the input image can be efficiently removed.
[0086]
In this embodiment, the minimum value filter is applied to the reference character string image and then binarized. However, the present invention is not limited to this, and the minimum value filter is applied. Alternatively, it can be binarized as it is. That is, the reason why the minimum value filter is applied is that the positional deviation of the input image is taken into consideration, so that when the positional deviation does not occur so much, the reference character string image can be binarized as it is.
[0087]
In this embodiment, the text character string input by the operator is registered in the specific area dictionary 105. However, the present invention is not limited to this, and character recognition is performed to specify the reference image. A character string included in the area can also be registered in the specific area dictionary 105.
[0088]
In the present embodiment, for the sake of convenience of explanation, the case where the character portion has a low density value (black) and the background portion has a high density value (white) has been described, but the present invention is not limited to this. Instead, it can also be applied to the case where the character portion has a high density value (white) and the background portion has a low density value (black). For example, when white letters are printed on a form, the character portion has a high density value (white) and the background portion has a low density value (black).
[0089]
Even in such a case, as described in detail in the above embodiment, a black pixel portion (background portion of a white eye) is applied by applying a minimum value filter to a binary image obtained by binarizing the reference character string image. A template in which each pixel of this image and each pixel forming a binary image obtained by binarizing the input character string image is a black pixel is created as a pixel, By cutting out the pixel value of the template portion from the input character string image and extracting the noise cut image from which the noise component has been removed from the input character string image, even a form including white characters can be efficiently and A form can be discriminated with high accuracy.
[0090]
【The invention's effect】
As described above in detail, according to the invention of claim 1, each pixel forming a binary image obtained by binarizing the reference image, and a pixel forming a binary image of the input image corresponding to the pixel, However, the pixel value of a pixel that is both a black pixel is defined as a black pixel. , Other pixel values are white pixels Generated templates, Performs die cutting with the pixel values of the pixels other than the portion corresponding to each black pixel of the template as white pixels in the input image Since the noise cut image is generated by removing the noise component from the input image, an image processing apparatus capable of efficiently removing noise is obtained.
[0091]
According to the second aspect of the present invention, the minimum value filter is applied to the reference image to expand the black pixel portion, and the image to which the minimum value filter is applied is binarized. There is an effect that an image processing apparatus capable of efficiently dealing with the deviation can be obtained.
[0092]
According to the invention of claim 3, among the pixels forming the noise cut image, each pixel that does not have a pixel value cut out from the input image is obtained from the pixel values of a plurality of pixels in the input image that are not to be cut out. Since the obtained value is configured to have the pixel value, there is an effect that an image processing apparatus capable of reconstructing the pixel value of the pixel that is not to be punched into an appropriate value is obtained.
[0093]
According to the invention of claim 4, the input image is invalid based on the difference between the number of black pixels of the binary image obtained by binarizing the input image and the number of black pixels of the binary image of the noise cut image. Since it is configured to determine whether or not it is an image, there is an effect that it is possible to obtain an image processing apparatus capable of efficiently preventing fraud in which the die removal using the template is taken.
[0094]
According to the invention of claim 5, each pixel forming a binary image obtained by binarizing the reference image and each pixel forming the binary image of the input image corresponding to the pixel are black pixels. The pixel value of a pixel is black , Other pixel values are white pixels Generated templates, Performs die cutting with the pixel values of the pixels other than the portion corresponding to each black pixel of the template as white pixels in the input image Since the noise cut image is generated by removing the noise component from the input image, an image processing method capable of efficiently removing the noise is obtained.
[0095]
According to the sixth aspect of the present invention, the minimum value filter is applied to the reference image to expand the black pixel portion, and the image to which the minimum value filter is applied is binarized. There is an effect that an image processing method capable of efficiently dealing with the deviation can be obtained.
[0096]
According to the invention of claim 7, among the pixels forming the noise cut image, each pixel having no pixel value cut out from the input image is determined from the pixel values of a plurality of pixels in the input image not to be cut out. Since the obtained value is used as the pixel value, there is an effect that an image processing method capable of reconstructing the pixel value of the pixel that is not to be punched into an appropriate value is obtained.
[0097]
According to the invention of claim 8, the input image is invalid based on the difference between the number of black pixels of the binary image obtained by binarizing the input image and the number of black pixels of the binary image of the noise cut image. Since it is configured to determine whether or not it is an image, there is an effect that an image processing method capable of efficiently preventing fraud in which die cutting using a template is taken in reverse is obtained.
[0098]
According to the ninth aspect of the present invention, since the computer program executes the method according to any one of the fifth to eighth aspects, the operation according to any one of the fourth to sixth aspects is performed by the computer. Can be realized.
[Brief description of the drawings]
FIG. 1 is a functional block diagram showing a configuration of a form discriminating apparatus used in an embodiment of the present invention.
FIG. 2 is an explanatory diagram for explaining a concept of ruled line feature extraction processing by a ruled line feature extraction unit shown in FIG. 1;
FIG. 3 is an explanatory diagram illustrating an example of ruled line feature extraction by the ruled line feature extraction unit illustrated in FIG. 1;
FIG. 4 is a flowchart showing a processing procedure when a form is registered as a dictionary for comparison at the time of discrimination.
FIG. 5 is a flowchart showing a form discrimination processing procedure performed by the form discrimination apparatus shown in FIG. 1;
FIG. 6 is a diagram illustrating an example of a form to be determined in the present embodiment.
7 is an explanatory diagram for explaining a form whose details are determined by a detail determination unit shown in FIG. 1; FIG.
8 is a functional block diagram showing a configuration of a dictionary creation unit shown in FIG.
FIG. 9 is a flowchart showing a processing procedure of a specific area dictionary creation unit shown in FIG. 8;
FIG. 10 is an explanatory diagram illustrating an example of a character string image.
11 is a functional block diagram illustrating a configuration of a detail determination unit illustrated in FIG. 1. FIG.
12 is a flowchart illustrating a processing procedure of a detail determination unit illustrated in FIG.
FIG. 13 is an explanatory diagram illustrating an example of a character string image.
[Explanation of symbols]
10 Form discrimination device
101 Image input unit
102 Ruled line feature extraction unit
103 Dictionary creation part
104 Ruled line feature dictionary
105 Specific area dictionary
106 Ruled line feature matching unit
107 Detailed judgment section
108 Output section
800 Ruled line feature dictionary creation part
810 Specific area dictionary creation part
811 Cutout processing part
812 Minimum value filtering unit
813 binarization processing unit
1101 Template creation part
1102 Noise cut image extraction unit
1103 Fraud determination part
1104 Character recognition processing unit
1105 Character determination part

Claims (9)

In an image processing apparatus for removing noise components included in the input image when comparing both images by comparing the input image with a reference image,
Each pixel that forms a binary image obtained by binarizing the reference image and a pixel that forms a binary image of the input image corresponding to the pixel are both black pixels, and the pixel value of the pixel is a black pixel . Template generating means for generating a template in which the pixel values of other pixels are white pixels ;
A noise cut image obtained by performing die cutting using a pixel value of a pixel other than a portion corresponding to each black pixel of the template generated by the template generation unit as a white pixel in the input image, and removing a noise component from the input image An image processing apparatus comprising: a noise-cut image generation unit that generates   The template generation unit applies a minimum value filter to the reference image to expand a black pixel portion, and binarizes an image to which the minimum value filter is applied by the minimum value filtering unit. The image processing apparatus according to claim 1, further comprising: means.   Among the pixels forming the noise cut image generated by the noise cut image generation means, each pixel having no pixel value cut out from the input image is determined from the pixel values of a plurality of pixels in the input image that are not to be cut out. The image processing apparatus according to claim 1, wherein the obtained value is a pixel value.   Based on the difference between the number of black pixels of the binary image obtained by binarizing the input image and the number of black pixels of the binary image of the noise cut image generated by the noise cut image generation unit, the input image is invalid. 4. The image processing apparatus according to claim 1, further comprising fraud determination means for determining whether or not the image is a correct image. In the image processing method for removing a noise component included in the input image when comparing both images by comparing the input image with a reference image,
Each pixel that forms a binary image obtained by binarizing the reference image and a pixel that forms a binary image of the input image corresponding to the pixel are both black pixels, and the pixel value of the pixel is a black pixel . A template generation step of generating a template in which the pixel values of other pixels are white pixels ;
A noise cut image obtained by performing die cutting with a pixel value of a pixel other than a portion corresponding to each black pixel of the template generated by the template generation step in the input image as a white pixel, and removing a noise component from the input image And a noise cut image generation step for generating the image processing method.   The template generation step includes a minimum value filtering step of expanding a black pixel portion by applying a minimum value filter to the reference image, and a binarization that binarizes an image to which the minimum value filter is applied by the minimum value filtering step. The image processing method according to claim 5, further comprising a step.   Among the pixels forming the noise cut image generated by the noise cut image generation step, each pixel that does not have a pixel value cut out from the input image is determined from pixel values of a plurality of pixels in the input image that are not to be cut out. The image processing method according to claim 5, wherein the obtained value is a pixel value.   Based on the difference between the number of black pixels of the binary image obtained by binarizing the input image and the number of black pixels of the binary image of the noise cut image generated by the noise cut image generation step, the input image is invalid. The image processing method according to claim 5, further comprising a fraud determination step of determining whether the image is a correct image.   The program which makes a computer perform the method as described in any one of the said Claims 5-8.

Images