JP2003346083A - Character recognition device and method, program, storage medium, and character recognition system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve a high recognition rate even for characters of high flatness by using a recognition dictionary for standard patterns with fixed vertical and horizontal dimensions. <P>SOLUTION: The aspect ratio R of a character block is determined and compared with predetermined thresholds T1, T2, T3 and T4. By comparing with the thresholds, the character block is determined to be either a character block within the range of standard aspect ratios or a flat (portrait or landscape) character block or a character block that could be both the character block with the standard aspect ratios or the flat character block. Then normalization suited for each of the blocks is performed and character recognition is carried out. <P>COPYRIGHT: (C)2004,JPO

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device, a character recognition method, a program and a storage medium, and a character recognition system for optically reading a document and recognizing characters. It concerns the character normalization to be performed.

[0002]

2. Description of the Related Art In an optical character recognition device (OCR), a character line is cut out (extracted) by density projection (histogram), and a character block is cut out (extracted) for each character.
I do. When extracting a character block, density projection is performed in the character line direction, character lines are separated based on changes in the density projection value, and density projection is performed for each character line in the direction perpendicular to the character line. Extract the character block of. If necessary, the final character block, which becomes a character image in units of one character, is cut using information such as a standard character size and character pitch estimation value and a density projection value in a direction perpendicular to the line. put out. The extracted character block is subjected to predetermined feature data extraction processing after normalizing the vertical and horizontal dimensions. For the character to be recognized, the feature data of the image is obtained in advance and registered as a standard pattern. Each character block from which feature data is extracted is
The similarity with the standard pattern is calculated, and the character with the highest similarity is regarded as the recognition result. A set of standard patterns is called a recognition dictionary.

The normalization of a character block is a process of setting the vertical and horizontal dimensions (normalized size) corresponding to the shape of a standard pattern. Conventionally, Japanese Patent Laid-Open No. Hei 7-271908 and Japanese Patent No. 3121
No. 091 and Japanese Patent No. 32169983 are known. (1) Processing described in JP-A-7-271908 (conventional example 1
That. According to the aspect ratio, the normalized size is divided into multiple stages, and the normalized size is normalized to the closest normalized size. (2) Processing described in Japanese Patent No. 3121091 (referred to as Conventional Example 2) If the aspect ratio is within a predetermined range, the character image is enlarged / reduced so as to match a predetermined vertical / horizontal dimension, and is normally formed into a square character frame. When the aspect ratio is larger or smaller than a predetermined range, that is, for a vertically long character or a horizontally long character, a normalized size capable of preserving the vertically and horizontally long characteristics is adopted. The lengthwise and widthwise character images each have an aspect ratio multiplied by a predetermined coefficient. (3) Processing described in Japanese Patent No. 3216983 (referred to as Conventional Example 3) Character images are divided into multiple stages according to the aspect ratio.
Scaling is performed according to each stage, and the normalized size is gradually changed.

[0004]

However, in the conventional examples 1 and 2,
In the same character type, the normalized size differs depending on the character.On the other hand, since each standard pattern in the recognition dictionary is set to a fixed vertical and horizontal dimension, the process of determining similarity becomes complicated and the recognition rate decreases. I do.

In Conventional Example 3, when the flatness of a character is remarkable, that is, when the degree of length and width becomes remarkable, deformation cannot be ignored and the recognition rate decreases.

The present invention has been made in order to solve such a conventional problem, and obtains a high recognition rate even for a character having a large flatness by using a recognition dictionary of a standard pattern having a constant vertical and horizontal dimension. The purpose is to:

[0007]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, a character recognition apparatus of the present invention extracts a character block from a recognition dictionary storing characteristic data of a character standard pattern and image data including a character image. Character block extracting means;
Calculating the aspect ratio of the extracted character block, using a normalization procedure determining means for determining a normalization procedure based on the aspect ratio, using a normalization procedure determined by the normalization procedure determination means, A character data obtaining unit that obtains the characteristic data of the character block; and a similarity calculating unit that calculates a similarity by comparing the characteristic data obtained by the characteristic data obtaining unit with the characteristic data of the standard pattern. In the recognition apparatus, the normalization procedure determining means converts the character block extracted by the character block extracting means into an aspect ratio (hereinafter, referred to as an aspect ratio (hereinafter, referred to as an aspect ratio) substantially equal to a standard pattern) based on the calculated aspect ratio. It is called "standard aspect ratio.")
Character block, and a character block that may be either a flat character block or a character block having a standard aspect ratio, which character block is determined, and is determined as the character block having the standard aspect ratio. The character block determined to use the normalization procedure for a character block having a standard aspect ratio is determined as a character block that may be either the flat character block or the character block having a standard aspect ratio. For the block, it is determined to use both the normalization procedure for the character block having the standard aspect ratio and the normalization procedure for the flat character block.

In order to solve the above problem, a character recognition method according to the present invention comprises: a character block extracting step of extracting a character block from image data including a character image; and calculating an aspect ratio of the extracted character block. A normalization procedure determining step of determining a normalization procedure based on the aspect ratio;
Using the normalization procedure determined in the normalization procedure determination step, a feature data obtaining step for obtaining the feature data of the character block, the feature data obtained in the feature data obtaining step, and stored in a recognition dictionary. A similarity calculation step of calculating a similarity by comparing with the characteristic data of the standard pattern, wherein the normalization procedure determination step includes the step of determining the character block based on the calculated aspect ratio. The character blocks extracted in the extraction step are divided into at least a character block having an aspect ratio substantially equal to the aspect ratio of the standard pattern (hereinafter referred to as “standard aspect ratio”), a flat character block and a character block having the standard aspect ratio. It is determined whether the character block is a character block having the above-mentioned standard aspect ratio. The determined character block is determined to use a normalization procedure for a standard aspect ratio character block, and is determined to be a character block that may be either the flat character block or the standard aspect ratio character block. For the character block, it is determined to use both the normalization procedure for the standard aspect ratio character block and the flat character block normalization procedure.

In order to solve the above-mentioned problems, a character recognition device for a character recognition system according to the present invention comprises: a communication unit capable of receiving image data including a character image; a recognition dictionary that can be referred to; A character block extracting unit for extracting a character block; a normalization procedure determining unit for calculating an aspect ratio of the extracted character block; and determining a normalization procedure based on the aspect ratio; and the normalization procedure determining unit. A feature data obtaining unit that obtains feature data of the character block using the normalization procedure determined by the above, and feature data of a standard pattern stored in the recognition dictionary and the feature data obtained by the feature data obtaining unit. A similarity calculating means for calculating a similarity by comparing the character recognition apparatus for a character recognition system, wherein the normalization procedure determining means comprises: Based on the calculated aspect ratio, the character block extracted by the character block extraction means is divided into a character block having an aspect ratio that is at least substantially equal to the aspect ratio of the standard pattern (hereinafter, referred to as “standard aspect ratio”) and a flat shape. Character block that is likely to be a character block having a normal aspect ratio or a character block having a standard aspect ratio. It is determined to use the normalization procedure for the character block having the aspect ratio, and the character block determined as a character block that may be either the flat character block or the character block having the standard aspect ratio is the standard aspect ratio. Deciding to use both the normalization procedure for the ratio character block and the normalization procedure for the flat character block, That.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, a character recognition device according to the present invention,
An embodiment of a character recognition method, a program and a storage medium, and a character recognition system will be described with reference to the drawings.

[Character Recognition Apparatus] FIG. 1 is a block diagram showing an embodiment of a character recognition apparatus such as an optical character recognition apparatus (OCR) according to the present invention.

The character recognition device includes an external storage device 104 such as a magnetic disk and an image scanner 108, and reads image data of an original from the external storage device 104 or directly reads an original with the image scanner 108.
The character recognition device includes a CPU 101 for overall control,
The control program of the CPU 101 is stored in the ROM 102. The image input from the image scanner 108 is held in the RAM 103, and a process for character recognition is executed by a control program in the ROM 102. Recognition dictionary for character recognition is ROM with control program
The data is stored in the RAM 102 or read from the external storage device 104 into the RAM 103 for each required character type. If necessary, preprocessing such as noise removal is performed prior to character recognition.

In FIG. 1, reference numeral 105 denotes a display,
06 is a keyboard, 107 is a pointing device such as a mouse. A network interface 109 communicates with a remote device (not shown) and can transmit and receive programs, recognition dictionaries, data, and the like.

Next, a character recognition method executed by the character recognition device will be described.

[Character Recognition Method] [First Embodiment] FIG.
Is the first character recognition process performed by the character recognition device of FIG.
It is a flowchart which shows embodiment.

In the first embodiment, the normalization method is changed according to the aspect ratio of the character image, and when it is not easy to determine the normalization method, the similarity is provisionally determined using a plurality of normalization methods. Calculate and determine the final normalization method based on the calculation result. In the present embodiment, the aspect ratio of the normalized size is 1: 1.

Step S201: First, an image read by the scanner 108 or an image stored in the external storage 104 is input.

Step S202: A character line is extracted from the input image. At the time of character line extraction, the character line direction is detected, the density projection in the character line direction is taken, and the density projection value (histogram) is obtained. Based on the change in the density projection value, a portion where the density projection value is equal to or more than a predetermined value is extracted as a character line.

Step S203: A character block of one character unit is extracted from the extracted character line. Step S20
A density projection is performed on each of the character lines extracted in step 2 in a direction perpendicular to the character line, and a portion where the density projection value is equal to or more than a predetermined value is extracted as a character block based on a change in the density projection value. Depending on the document, each character block may be a character image in units of one character. However, in a document having a small character pitch, one character may be divided. The density projection value in the vertical direction with respect to the character line is evaluated while estimating the character pitch and the character pitch, and control is performed so that a character image in units of one character is extracted as a final character block.

Step S204: The aspect ratio of the extracted character block is determined, and predetermined threshold values T1, T2, T3, T4
Compare with Here, in the character block 301 in FIG. 3 (showing an image of one character of Hiragana), when the height is h and the width is w, an aspect ratio R = w / h is defined. The threshold values T1, T2, T3, and T4 are T1 <T2 <T3 <
When R is smaller than T1 (Equation 1), the character block is vertically long, and when R is larger than T4 (Equation 2), the character block is horizontal, R is T2 or more, and T3 or less. At this time (Equation 3), it is determined that the character block has an aspect ratio substantially equal to the aspect ratio of the standard pattern (hereinafter, referred to as “standard aspect ratio”).

[0021]

(Equation 1) If the character block has the standard aspect ratio, step S2
The process proceeds to step S206, and if it is portrait or landscape, the process proceeds to step S206.

When the aspect ratio R is equal to or greater than T1 and smaller than T2 (Equation 4), the character block is determined to be a portrait or a standard aspect ratio, and the final determination of which is the character block is suspended. When the aspect ratio R is greater than T3 and equal to or less than T4 (Equation 5), the character block is determined to be landscape or standard aspect ratio, and the final determination as to which is the character block is suspended.
Thereafter, the process proceeds to step S208.

[0023]

(Equation 2)

For character originals containing kanji, the threshold values are, for example, T1 = 0.4, T2 = 0.6, T3 = 1.4, T4
= 2.5, a good recognition result is obtained.

The character block 301 shown in FIG.
2.0, which is processed in step S208.

Step S205: Since the character block is within the range of the standard aspect ratio, normalization is performed so that the vertical and horizontal dimensions of the character block become the normalized size. In this case, the character block may be slightly deformed, but there is no problem because the character block is deformed to such an extent that the recognition rate is not adversely affected.

Step S206: For a vertically long or horizontally long character block, the longer length, that is, the height of a vertically long character block, and the width of a horizontally long character block are matched with the normalized size, and the aspect ratio R Determine the other length to hold As a result, the characteristics of the character's aspect ratio R are preserved, and a reduction in the recognition rate due to the character block deformation can be prevented. Note that, for the normalized size, the vertical characters are not enough left and right, and the horizontal characters are not enough up and down.
Left and right margins or upper and lower margins are added so that the character is located at the center, and the characters are made to match the normalized size so that good recognition results can be obtained.

Step S208: When it is determined that the character block has the portrait or standard aspect ratio, or when it is determined that the character block has the landscape or standard aspect ratio, the process proceeds to step S208.
Both normalization processing similar to the normalization processing of the standard aspect ratio in 205 and normalization processing similar to the normalization processing of the vertical character block (or the horizontal character block) in step S206 are executed.

For example, the character block 301 (R = R
2.0), the normalization process for the standard aspect ratio (the normalization result is shown by the normalized image 302 in FIG. 3A).
And a normalization process for the horizontally long character block (the normalization result is shown by a normalized image 303 in FIG. 3B).

Step S207: When the normalization processing in step S205 or step S206 is completed, in step S207, the image after the normalization is identified and calculated. In the identification operation, feature data is extracted from the normalized image, the similarity with the standard pattern for each character type stored in the recognition dictionary is calculated, and the process proceeds to step S211 to determine the character having the highest similarity. And

Step S209: If two normalized images are obtained by the normalization processing in step S208, step S209
In step 209, both normalized images are identified and calculated, and in step S2
Move to 10.

Step S210: Step S209-2
The two identification calculation results are compared, and the normalized result with the higher similarity and the similarity are adopted, and the process proceeds to step S211.

Step S211: The character most similar to the character block is determined based on the results of the normalization selection and the identification operation.

In the above-described character recognition processing, two normalization methods may be executed. However, when a recognition dictionary used for the character recognition processing is created, there are a plurality of normalization methods. In this embodiment, when a recognition dictionary is created, a certain normalization method is selected for each character type (that is, the same character is used) because the configuration of the dictionary becomes significantly complicated or the standard pattern cannot be determined uniformly. A recognition dictionary is created using a plurality of learning character image data serving as codes, but the same normalization method is selected for learning data having the same character code. When selecting the normalization method, an aspect ratio close to the mode value or the average value of the aspect ratio in the learning data of the recognition dictionary can be adopted for each character type. It is also possible to determine the aspect ratio based on a rule of thumb.

As described above, in the first embodiment,
Since an optimal normalization method is selected for each character block in consideration of the degree of similarity with the recognition dictionary, a high recognition rate can be ensured even if a recognition dictionary having a standard pattern with a fixed aspect ratio is used. [Second Embodiment] FIG. 4 is a flowchart showing a second embodiment of the character recognition method executed by the character recognition device of FIG.

The second embodiment normalizes feature data instead of the character block itself, and changes the normalization method according to the aspect ratio of the character image. The similarity is calculated using a plurality of normalization methods, and the final normalization method is determined based on the calculation results.

Step S401: An image is input as in step S201 of the first embodiment.

Step S402: Similar to step S202 of the first embodiment, a character line is extracted.

Step S403: As in step S203 of the first embodiment, a character block is extracted.

Step S404: The aspect ratio R of the extracted character block is determined, and the aspect ratio R is set to a predetermined threshold value T1, T2.
(T1 <T2). When R is equal to or greater than T1 and equal to or less than T2 (Equation 6), it is determined that the character block is within the range of the standard aspect ratio. When R is smaller than T1 (Equation 7), it is determined that the character block belongs to either the portrait or standard aspect ratio. When R is larger than T2 (Equation 8),
It is determined that the character block belongs to either the landscape or the standard aspect ratio. If it is determined that the character block has the standard aspect ratio (T1 ≦ R ≦ T2), the process proceeds to step S405. If the character block is portrait (R <T1), the process proceeds to step S408. If it is landscape (R> T2), the process proceeds to step S409. Transition.
For a character document containing kanji, the threshold value is set to, for example,
Good recognition results are obtained by setting 0.4 and T2 = 2.5.

[0041]

(Equation 3) Step S405: The contour of the image in the character block is extracted, and the characteristic data of the character block is extracted by giving the direction index of FIG. 8 to the boundary pixels forming the contour. The boundary pixels in the character block are evaluated in the order in which the scanning from left to right proceeds sequentially from top to bottom, and the direction of the pixel following the boundary pixel P0 of interest is the four non-opposite directions among the eight pixels surrounding the pixel of interest. (Eg, lower, lower right, right, upper right) is sufficient to express the feature data. These directions are, for example, C1
(Lower), C2 (lower right), C3 (right), C4 (upper right) are represented by directional indices.

As shown in FIG. 7, a character block 701
Assuming that the width of (a single character image of Hiragana "i") is w and the height is h, the character block 701 divides the width and height into a predetermined number to form a plurality of small areas. The number of divisions is, for example, n = 4 in the width direction and m = 4 in the height direction.

In FIG. 7, one small area 702 is shown as a representative, and among the pixels in the small area, boundary pixels are shaded, and P1, P2, P3, P4, P5, P6, P7, P8 Is given. When these boundary pixels are viewed in the scanning order from left to right and from top to bottom, a pixel P2 follows the pixel P1, and a pixel P3 follows the pixel P2. Further, a pixel P5 exists following the pixel P4, a pixel P6 exists following the pixel P5, a pixel P7 exists following the pixel P6,
Following the pixel P7, there is a pixel P8. In this continuous state, the direction index C2 is applied to the pixel P1, and the direction index C
2. Direction index C2 for pixel P4, direction index C for pixel P5
3, the direction index C2 for the pixel P6 and the direction index C2 for the pixel P7
Is expressed by giving

In each small area, the direction index is statistically processed,
For example, the number of each direction index is set as the feature data of the small area. Since C2 occurs five times and C3 occurs once in the small area 702, if the feature data is {C1, C2, C3, C4}, it will be {0, 5, 1, 0}. Alternatively, if the most frequent direction index is the feature data of the small area, the global direction of the image is expressed.

Step S406: The feature data of the entire character block is calculated while normalizing the feature data of each small area obtained in step S405. Assuming that the normalized size is N × N as in FIG. 3, the character block is enlarged (N / w) times in the horizontal direction and (N / h) times in the vertical direction.
By biasing the feature data according to the enlargement ratio, the feature data can be normalized. Regarding the direction index in the oblique direction, a square having the same area as the character block 701 is assumed, and the ratio between the square and the normalized size is multiplied. That is, Equation 9, Equation 10, Equation 1 are assigned to each direction index.
1. The normalized feature data Cn1, Cn2, Cn3, and Cn4 are calculated by multiplying the coefficients shown in Equation 12 by 1.

[0046]

(Equation 4) Step S407: A character block identification calculation is performed based on the feature data normalized in step S406.
The discrimination calculation calculates the similarity between the normalized feature data and the standard pattern for each character type stored in the recognition dictionary, and proceeds to step S414 to determine the character having the highest similarity as the recognition result.

Step S408: When the aspect ratio R is smaller than T1, the character block is determined to be either the portrait or the standard aspect ratio, and the final decision as to which is the character block is suspended. As shown in FIG. 6, when there is a possibility that the character block a is vertically long, a character block b is generated by adding a margin corresponding to the character block height h to the left and right of the character block a. The width of each margin is (h × T1-w) / 2,
As a result, the width of the character block is h × T1. This is the minimum width that is within the standard aspect ratio range at height h,
The character block after adding the margin has the aspect ratio R = T1. Thereafter, the process proceeds to step S410.

Step S409: When the aspect ratio R is larger than T2, the character block is determined to be either the landscape or the standard aspect ratio, and the final determination as to which is the character block is suspended. As shown in FIG. 5, when there is a possibility that the character block a is horizontally long, a character block b is generated by adding a margin corresponding to the character block width w above and below the character block a. The height of each margin is (w / T2-h) / 2,
As a result, the width of the character block b becomes w / T2. This is the minimum height within the standard aspect ratio range when the width is w, and the character block after adding the margin has the aspect ratio R = T2. Thereafter, the process proceeds to step S410.

Step S410: Step S408, S
At 409, feature data is extracted for both the character block to which margin addition processing has been performed (character block b in FIGS. 5 and 6) and the character block to which no margin has been added (character block a in FIGS. 5 and 6). . The process of extracting characteristic data is the same as in step S405.
Thereafter, the process proceeds to step S411.

Step S411: Normalize the characteristic data for both the character block (character block b in FIGS. 5 and 6) subjected to the margin addition processing and the character block without the margin addition (character block a in FIGS. 5 and 6). Become The normalization process is the same as in step S406. Thereafter, the process proceeds to step S412.

Step S412: Identification calculation is performed for both the character block subjected to the margin addition processing (character block b in FIGS. 5 and 6) and the character block without margin addition (character block a in FIGS. 5 and 6). I do.
The identification calculation is the same as in step S407. Thereafter, the flow shifts to step S413.

Step S413: The result of the discrimination calculation is performed on both the character block subjected to the margin addition processing (the character block b in FIGS. 5 and 6) and the character block without the margin addition (the character block a in FIGS. 5 and 6). That is, the similarities are compared, and a character block having a higher similarity is adopted.

In steps S408 and S409, the judgment on the aspect ratio has been suspended, but a final judgment is made at this stage, and the process proceeds to step S414.

Step S414: A character recognition result is determined based on the result of the identification calculation calculated in step S407 or selected in step S413.

In the second embodiment, since the character block image itself is not normalized, there is no adverse effect of noise component amplification, such as enlargement of fine irregularities in the outline of the image in the character block.

Also, by adding the minimum necessary margins to the vertically and horizontally long character blocks, the same processing as for the character blocks having the standard aspect ratio can be performed.
It is possible to use a recognition dictionary composed of standard patterns having a constant vertical and horizontal dimension. Therefore, the recognition rate does not decrease due to the character block deformation.

In the second embodiment, step S40
In 4, the determination is made by using the predetermined threshold values T1 and T2 separately for character blocks within the standard aspect ratio range and character blocks belonging to either the flat (vertical or horizontal) or standard aspect ratio. In the same manner as in the first embodiment, using the four threshold values T1, T2, T3, and T4, a character block within standard vertical and horizontal criticism, a flat (vertical or horizontal) character block, and a flat (vertical or horizontal) Landscape)
Alternatively, the processing may be performed separately for character blocks belonging to any of the standard aspect ratios. In this case, if it is determined that the block is a flat character block, the margin addition processing is performed as described above to extract the characteristic data, and the characteristic data is further normalized before performing the identification operation.

Note that the feature data in the second embodiment
The normalization is an example, and the image density or the like may be used as the feature data. The number of divisions, the division method, and the normalization method of the character block are not limited thereto.
For example, in the example described above, the direction index is C1 (lower),
C2 (lower right), C3 (right) and C4 (upper right) were used, but C1 (lower left), C2 (lower), C3 (lower right), C4
The evaluation may be performed using a direction index such as (right).

[Program] The control program in the character recognition apparatus shown in FIG. 1 includes a program for executing the first and / or second embodiment of the character recognition method. The program for implementing the first embodiment includes a routine for executing the processing of steps S201 to S211 and the program for implementing the second embodiment includes a routine for executing the processing of steps S401 to S414.

[Storage Medium] The character recognition method according to the present invention can be implemented by a device other than the character recognition device shown in FIG. 1, for example, a general-purpose computer. It may be set in an external storage device of a computer and executed by a CPU or the like in accordance with an instruction from an operator or the like.

FIG. 9 is a memory map showing a storage medium storing a program for executing the first and second embodiments of the character recognition method.

In FIG. 9, directory information is stored in addresses A0 to A1, a program for executing the first embodiment is stored in an area from an address next to A1 to address A2, and a program next to A2 is stored. A program for executing the second embodiment is stored in an area from an address to an address A3.

Although not shown, data dependent on various programs may be managed as directory information, and a program for installing various programs in a computer may be stored in the directory information.

[Character Recognition System] FIG. 10 is a block diagram showing an embodiment of a character recognition system in which the character recognition method according to the present invention is executed by a plurality of devices.

In this embodiment, in the Internet service provider, a program including a recognition dictionary for character recognition is stored in the server SV1. The Internet service provider is connected to the Internet via communication means, and a plurality of users can use the character recognition program by accessing the server SV1 from the Internet.

The user has a scanner SC1 and a personal computer PC1, and transmits an image of a document read by the scanner SC1 to the server SV1. The server executes character recognition processing using the image of the document as an input image.

Note that the recognition dictionary does not necessarily need to be present on the same computer as the character recognition program, and it is sufficient that the recognition dictionary can be referred to via communication means or the like. For example, the character recognition program may be provided on the server SV1 and the recognition dictionary may be provided on the separate server SV2, and may be provided as separate services. In this case, each of the holder of the character recognition program and the holder of the recognition dictionary can hold intellectual assets such as programs and dictionaries therein, thereby preventing illegal copying and the like.

[0068]

According to the present invention, a high recognition rate can be obtained even for a character having a large flatness by using a recognition dictionary of a standard pattern having a constant vertical and horizontal dimension.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a character recognition device according to the present invention.

FIG. 2 is a flowchart illustrating a first embodiment of a character recognition method performed by the character recognition device of FIG. 1;

FIG. 3 is a conceptual diagram showing an outline of normalization in the character recognition method of FIG.

FIG. 4 is a flowchart illustrating a second embodiment of a character recognition method performed by the character recognition device of FIG. 1;

FIG. 5 is a conceptual diagram showing details of a method for normalizing a horizontally long character in the character recognition method of FIG. 4;

FIG. 6 is a conceptual diagram showing details of a method for normalizing vertically long characters in the character recognition method of FIG. 4;

FIG. 7 is a conceptual diagram showing feature amounts in the character recognition method of FIG.

FIG. 8 is a conceptual diagram showing a direction index for setting the feature amount in FIG. 7;

9 is a conceptual diagram showing a memory map of a storage medium that stores various data processing programs executed by the character recognition device of FIG.

FIG. 10 is a block diagram showing a character recognition system in which a character recognition method according to the present invention is executed by a plurality of devices.

[Explanation of symbols]

101 ... CPU 102 ROM 103 RAM 104: External storage device 105 ・ ・ ・ Display 106 ... keyboard 107 Pointing device 108 ・ ・ ・ Image scanner 109 ・ ・ ・ Network interface

Claims (16)

[Claims] 1. A recognition dictionary storing characteristic data of a standard pattern of characters, a character block extracting means for extracting a character block from image data including a character image, and calculating an aspect ratio of the extracted character block; A normalization procedure determining unit that determines a normalization procedure based on the aspect ratio; and a feature data obtaining unit that obtains feature data of the character block using the normalization procedure determined by the normalization procedure determination unit. A similarity calculating unit that calculates a similarity by comparing the feature data obtained by the feature data obtaining unit with the feature data of the standard pattern, wherein the normalization procedure determining unit includes: Based on the calculated aspect ratio, the character block extracted by the character block extracting means is converted into an aspect ratio at least substantially equal to the aspect ratio of the standard pattern (hereinafter, referred to as the standard pattern). It is determined whether the character block is a character block of a character block having a standard aspect ratio or a flat character block or a character block having a standard aspect ratio. For the character block determined as the character block of the aspect ratio, it is determined to use the normalization procedure for the character block of the standard aspect ratio, and the possibility of both the flat character block and the character block of the standard aspect ratio is determined. A character recognition apparatus characterized in that a character block determined as a certain character block is determined to use both the normalization procedure for the standard aspect ratio character block and the normalization procedure for a flat character block. . 2. The normalization procedure determining unit determines a character block extracted by the character block extracting unit based on the calculated aspect ratio as a character block having a standard aspect ratio,
Judge which character block is a flat character block, or a character block that can be either a flat character block or a character block having a standard aspect ratio, and judge it as a character block having the standard aspect ratio. The determined character block is determined to use a normalization procedure for a character block having a standard aspect ratio, and the character block determined to be a flat character block is determined to use a normalization procedure for a flat character block. The character block which is determined as a character block which may be either the flat character block or the standard aspect ratio character block is determined by the normalization procedure for the standard aspect ratio character block and the flat character block. 2. The character recognition apparatus according to claim 1, wherein the determination is made to use both of the normalization procedures for use. 3. The feature data obtaining means extracts the feature data after normalizing the character block using the normalization procedure determined by the normalization procedure determination means, thereby obtaining the feature of the character block. 3. The character recognition device according to claim 1, wherein data is obtained. 4. The feature data obtaining unit extracts the feature data from the character block using the normalization procedure determined by the normalization procedure determination unit, and then normalizes the feature data, thereby obtaining the feature data. 3. The character recognition device according to claim 1, wherein characteristic data of the character block is obtained. 5. The feature data acquiring means for acquiring feature data using a normalization procedure for a character block having a standard aspect ratio means that a character block is deformed so that an aspect ratio matches a standard pattern. Is to obtain the feature data by extracting the feature data from the normalized character block image obtained by the above, and obtaining the feature data by using the normalization procedure for the flat character block, Normalization obtained by matching the longer side of the vertical or horizontal length to the corresponding side length of the standard pattern while maintaining the aspect ratio of, and adding a margin so that the shorter side becomes the normalized size 4. The character recognition device according to claim 3, wherein the characteristic data is obtained by extracting characteristic data from the character block image. Sense device. 6. The feature data acquiring means for acquiring feature data using a normalization procedure for a character block having a standard aspect ratio includes extracting feature data from the character block and extracting the extracted feature data. Acquiring the feature data by normalizing, and acquiring the feature data using the normalization procedure for the flat character block is to add a predetermined margin to the short side of the character block. 5. The character recognition apparatus according to claim 4, wherein the characteristic data is obtained by extracting characteristic data from the character block to which the margin has been added, and then normalizing the extracted characteristic data. 7. The normalization of the feature data is to apply a bias to the feature data extracted from the character block based on the ratio between the vertical and horizontal dimensions of the character block and the vertical and horizontal dimensions of the standard pattern. The character recognition device according to claim 4, wherein 8. The apparatus according to claim 1, wherein the feature data is a set of feature data obtained by dividing a character block into a plurality of small areas and obtaining the data for each of the small areas. Character recognition device. 9. The character recognition device according to claim 1, wherein the feature data is obtained by expressing a continuous state of boundary pixels of a character outline by a direction index. 10. A character block extracting step of extracting a character block from image data including a character image, calculating an aspect ratio of the extracted character block, and determining a normalization procedure based on the aspect ratio. A feature data obtaining step of obtaining feature data of the character block using the normalization procedure determined in the normalization procedure determining step; a feature data obtained in the feature data obtaining step; A similarity calculation step of calculating a similarity by comparing with the feature data of the standard pattern stored in the dictionary, wherein the normalization procedure determining step includes: The character block extracted in the character block extraction step is based on the aspect ratio (at least approximately equal to the aspect ratio of the standard pattern). Below, referred to as a “standard aspect ratio”), and a character block that can be either a flat character block or a character block with a standard aspect ratio. For the character block determined as the character block having the standard aspect ratio, it is determined to use the normalization procedure for the character block having the standard aspect ratio, and either the flat character block or the character block having the standard aspect ratio is possible. A character block determined as having a characteristic character block is determined to use both the normalization procedure for the character block having the standard aspect ratio and the normalization procedure for the flat character block. Recognition method. 11. In the normalization procedure determining step, the character block extracted in the character block extracting step is converted into a character block having a standard aspect ratio, a flat character block, and a flat character block based on the calculated aspect ratio. Character block that is likely to be a character block having a normal aspect ratio or a character block having a standard aspect ratio. Determined to use the normalization procedure for the character block of the aspect ratio, for the character block determined as the flat character block, determined to use the normalization procedure for the flat character block,
For the character block determined as a character block that may be either the flat character block or the standard aspect ratio character block, the normalization procedure for the standard aspect ratio character block and the normalization procedure for the flat character block The character recognition method according to claim 10, wherein a determination is made to use both of the conversion procedures. 12. The feature data obtaining step includes: extracting the feature data after normalizing the character block by using the normalization procedure determined in the normalization procedure determination step; 12. The character recognition method according to claim 10, wherein data is obtained. 13. In the feature data obtaining step, the feature data is extracted from the character block using the normalization procedure determined in the normalization procedure determination step, and then the feature data is normalized. 12. The character recognition method according to claim 10, wherein characteristic data of a character block is obtained. 14. A computer-executable program having a program code for causing a computer to execute each step of the character recognition method according to claim 10. 15. A storage medium storing the program according to claim 14. 16. A communication unit capable of receiving image data including a character image, a reference recognition dictionary, a character block extraction unit for extracting a character block from the received image data, and a vertical and horizontal direction of the extracted character block. A normalization procedure determining unit that calculates a ratio and determines a normalization procedure based on the aspect ratio; and using the normalization procedure determined by the normalization procedure determination unit, obtains the characteristic data of the character block. Character data comprising: a characteristic data obtaining unit; and a similarity calculating unit configured to calculate a similarity by comparing the characteristic data obtained by the characteristic data obtaining unit with the characteristic data of a standard pattern stored in the recognition dictionary. A character recognition device for a recognition system, wherein the normalization procedure determination means includes a sentence extracted by the character block extraction means based on the calculated aspect ratio. A character block may be a character block having at least an aspect ratio substantially equal to the aspect ratio of a standard pattern (hereinafter referred to as a “standard aspect ratio”), a flat character block and a character block having a standard aspect ratio. It is determined which of the character blocks is a character block, and a character block determined as the character block having the standard aspect ratio is determined to use a normalization procedure for a character block having a standard aspect ratio. For the character block determined as a character block that may be either the flat character block or the standard aspect ratio character block, the normalization procedure for the standard aspect ratio character block and the normalization procedure for the flat character block Character recognition device for a character recognition system, characterized in that both are determined to be used.