JPH0950489A - Character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character recognizing device which is capable of preventing the lowering of the recognition rate caused by the difference of the kind of a writing tool entering character in a medium to be read. SOLUTION: In the character recognizing device provided with a writing tool discrimination part 2, a feature extraction part 3 and a discrimination means 4, the feature extraction part 3 has a constitution that feature amt. to be extracted is made to differ according to the discrimination result obtained by the writing tool discrimination part 2. In the character recognizing device provided with the writing tool discrimination part, the feature extraction part, a blurring processing part and the discrimination part, the blurring processing part has a constitution that the degree of the blurring processing to be applied to the feature amt. of a character pattern is made to differ according the discrimination result obtained by the writing tool discrimination part.

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 for reading characters from an image obtained by optically scanning a medium to be read in which characters, symbols and the like (hereinafter referred to as "characters") are written. Is.

[0002]

2. Description of the Related Art A writing instrument used for writing characters on a medium such as paper or a form is generally a black pencil or a ballpoint pen. However, even if the letters are written in the same black,
Characters written with a pencil and those written with a ballpoint pen differ in the density of characters, the state of the edges of characters, the thickness of characters, etc. Therefore, when common character recognition processing is performed for both characters, they are recognized. Often negatively impacts rates. In order to avoid this, for example, in Japanese Unexamined Patent Publication No. 62-112124, a mark provided at a predetermined position on a medium to be read is filled with the same writing tool used for writing characters, and the filled mark is used. Reading with a sensor, comparing the output of this sensor with a reference value prepared in advance for each type of writing instrument to determine the writing instrument used, and to read the character pattern via an optical filter selected according to this determination , An optical character reader is disclosed.

[0003]

However, this conventional optical character reading apparatus uses only a suitable optical filter for image information, as in reading the image information on the medium to be read through the selected optical filter. It was a device that took in the information, and did not mention any subsequent information processing. Therefore, the device is not always one that can obtain good recognition accuracy.

[0004]

Therefore, in the first invention of this application, a writing tool discriminating section for discriminating the type of writing instrument used when writing characters on a medium to be read based on image data of the medium to be read. A preprocessing unit for obtaining a character pattern from the image data, a feature extraction unit for extracting a feature amount of the character pattern, and a plurality of standard character patterns prepared in advance for the extracted feature amount of the character pattern. A character recognition device for identifying the character pattern by comparing with the feature amount of the above, the feature extracting unit is configured to vary the feature amount to be extracted according to the determination result obtained by the writing instrument determining unit. It is characterized by being as.

In the first aspect of the present invention, differentiating the characteristic amount to be extracted specifically means differentiating the characteristic amount so as to absorb deformation such as blurring or crushing caused by a difference in the type of writing instrument. That is, the feature amounts are made different so that a stable feature amount corresponding to the feature amount of the standard character pattern can be obtained regardless of the type of writing instrument.

In the second invention of this application, a writing tool discriminating section for discriminating the kind of writing instrument used when a character is written on the medium to be read based on the image data of the medium to be read, and a character from the image data. A pre-processing unit for obtaining a pattern, a feature extraction unit for extracting a feature amount of the character pattern, a blur processing unit for performing a blur process on the extracted feature amount of the character pattern, and a blur processing unit for the blur process of the character pattern. In a character recognition device, comprising: an identification unit that identifies the character pattern by collating the completed feature amount with the feature amounts of a plurality of standard character patterns that are prepared in advance, the blur processing unit is the writing instrument determination unit. It is characterized in that the degree of blurring processing to be applied to the feature amount of the character pattern is varied according to the discrimination result obtained in (1).

Here, to make the degree of blurring different is specifically, the blurring is performed on the feature amount so as to absorb the deformation such as blurring or crushing caused by the difference in the type of writing instrument. That is, the blurring process is performed so that a stable feature amount corresponding to the feature amount of the standard character pattern can be obtained regardless of the type of writing instrument.

[0008]

According to the constitution of the first invention of this application, the feature extraction processing capable of removing the blur and the crushing due to the difference in the type of the writing implement can be performed, and according to the constitution of the second invention of the writing implement. It is possible to perform blurring processing that can remove blurring or crushing caused by a difference in type. Therefore, in any of the inventions, it is possible to obtain a character recognizing device that can obtain a feature amount from which blurring or crushing due to a difference in writing instrument type is removed and that can perform recognition processing using the feature amount.

[0009]

Embodiments of the present invention will be described below with reference to the accompanying drawings. However, the drawings used in the description are schematically shown to the extent that the present invention can be understood. In each of the drawings used for description, the same components are denoted by the same reference numerals, and overlapping description may be omitted.

[0010] 1. Description of First Invention FIG. 1 is a block diagram showing a configuration of a character recognition device according to an embodiment of the first invention. The character recognition device according to this embodiment includes a preprocessing unit 1, a writing instrument discrimination unit 2, and a feature extraction unit 3.
And an identification unit 4.

Here, the pre-processing unit 1 may have any suitable configuration, but in the case of this embodiment, FIG.
It has the configuration shown in. That is, the photoelectric conversion unit 1 for photoelectrically converting an optical signal obtained by scanning the medium to be read.
1, a first pattern register 12 for storing a multi-valued image obtained through the photoelectric conversion unit 11, and a binary patterning process for the multi-valued image stored in the first pattern register 12. It is composed of a binarization unit 13. In addition, in this embodiment, the writing instrument determination unit 2 uses R (red) and G (green).
Considering an example in which the writing tool is discriminated (details will be described later) based on pixel data of three colors B and (B), the photoelectric conversion unit 11 in the preprocessing unit 1
The density information for each of R, G, and B can be read.
The pattern register 12 has R (red) and G per pixel.
The density information of three colors (green) and B (blue) can be stored. Further, this first pattern register 12
Is capable of virtually setting XY coordinates on it and reading out pixel data at an arbitrary position represented by this coordinate system.

In the color image of this embodiment, black is used when the density information of R, G and B is all near 0.

Further, the writing instrument discriminating section 2 in the present embodiment has the first pattern register 1 as image data.
The character line pixels are extracted from the color image stored in No. 2 and the writing instrument is discriminated based on the difference in the R, G, and B value densities of the extracted pixels (the detailed description will be given. This will be described later in the description of the operation.). But,
The configuration of the writing instrument discriminating unit 2 is not particularly limited as long as the writing instrument can be discriminated and the result can be output to the feature extracting unit 3. For example, a structure using a filled mark as disclosed in JP-A-62-112124 may be used.

Further, in this embodiment, the feature extraction unit 3 has a second pattern register 31, a line width calculation unit 32, a character frame detection unit 33, and four sub-pattern extraction units, as shown in FIG. It is composed of units 34, 35, 36 and 37 and a feature matrix extraction unit 38. A detailed description of each of these components will be given later in the description of the operation, and the outline thereof is as follows.

First, the second pattern register 31 stores a character pattern binarized by the binarizing unit 13 of the preprocessing unit 1 (hereinafter referred to as "character pattern" for convenience of description). It is a thing. The binarization unit 13 uses R, G, and R for the multivalued image stored in the first pattern register 12 based on the format information stored in advance.
An appropriate threshold value is provided for each of B, and if all the RGB values of each pixel are equal to or less than the threshold value, a black pixel (1) is set, and if not, a white pixel (0) is set and binarization is performed. Similar to the first pattern register 12, the second pattern register 31 is capable of virtually setting XY coordinates on the second pattern register 31 and reading out pixel data at an arbitrary position represented by this coordinate system. There is. The line width calculator 32 is for detecting the line width from the character pattern stored in the second pattern register 31. Further, the character frame detection unit 33 is for detecting the character frame of the character pattern stored in the second pattern register 31. The sub-pattern extracting units 34 to 37 scan the character pattern stored in the second pattern register 31 in a plurality of directions and extract a plurality of sub-patterns representing stroke components for each scanning direction component. In this example, the sub-pattern extraction unit is configured for each of the vertical, horizontal, right diagonal, and left diagonal scanning directions. However, in this embodiment, each sub-pattern extraction unit, the details of which will be described later in the description of the operation, determines the extraction amount of the stroke component (specifically, the amount of black bits in the character pattern) for each direction component with a writing instrument. The configuration is adjusted according to the determination result of the determination unit 2. By doing so, the feature amount (feature matrix in this case) extracted by the feature extraction unit 3 is made different according to the determination result of the writing instrument determination unit 2. Further, the feature matrix extraction unit 38 sets the area corresponding to the character pattern stored in the second pattern register 31 for each sub-pattern m × n (m and n are integers of 1 or more, including m = n). And a line width w detected by the line width calculation unit 32 is used to extract a feature matrix as a feature amount. is there.

Next, the character recognition apparatus according to the embodiment of the first invention will be further described together with its operation. In addition,
This description will be given while considering an example in which character recognition is performed on the medium 41 to be read as shown in FIG. That is, the character itself is white, and a part of the surface thereof has a character entry frame 42 preliminarily printed in a color other than white and black, for example, orange, and the katakana “e” is included in the character entry frame 42. Is a read medium 41 written with either a ballpoint pen or a pencil as a writing tool.

First, the medium 41 to be read is scanned to obtain a color image (also referred to as a multi-valued image) having color information as image data from the medium 41 to be read. This can be performed by various known methods, for example, by the method described below with reference to FIG. By optically scanning the medium to be read 41 in the main scanning direction 51 and the sub-scanning direction 52 by a color scanner or the like, an image having pixel data of R, G, and B colors is sequentially loaded into the first pattern register 12. , A multi-valued image is input to the first pattern register 12.

When a multi-valued image is input to the first pattern register 12, the writing instrument discriminating section 2 extracts pixels constituting a character from the first pattern register 12, and further, R of the extracted pixels, The writing instrument is discriminated based on the G and B values. Extraction of pixels that make up these characters and R,
In this embodiment, the writing instrument is discriminated based on the G and B values as described below. First, the extraction of pixels forming a character is performed by utilizing the fact that the colors of the character 43, the character entry frame 42, and the medium to be read 41 are different from each other, such as black, orange, and white, respectively. That is, R, G, which the pixels forming the portion of the character 43 written in black have,
Generally, the B value is the portion of the character entry frame 42 or the medium 4 to be read.
Since all of the pixels forming part 1 have smaller values than those of the pixels, the pixels whose R, G, and B values are all smaller than the predetermined threshold value are extracted as the pixels forming the character.
In addition, the type of writing instrument, that is, whether the character is written by a ballpoint pen or a pencil here is determined as follows. First, a coordinate system in which the abscissa represents the concentrations of R, G, and B and the ordinate represents the frequency (hereinafter, “concentration-frequency coordinate system”).
That)). Then, paying attention to the respective densities of R, G, and B with respect to the pixels extracted as the pixels constituting the character, 1 is accumulated in the appearance frequency of those densities in the density-frequency coordinate system. When the frequency histogram is created in this manner, a concentration distribution having three peaks of R, G, and B can be obtained. This concentration distribution is different for a character written with a pencil and a character written with a ballpoint pen. Therefore, it is determined whether the writing instrument is a pencil or a ballpoint pen based on the difference in the density distribution. The method for extracting the pixels forming a character based on the R, G, B values and the above method for determining the type of writing instrument are described in detail in Japanese Patent Application No. 6-209805 filed by the applicant of the present application.

When the writing implement discriminating unit 2 completes the discrimination of the writing implement, the feature extracting unit 3 receives the type of the writing implement which is the discrimination result. In addition, the binarization unit 13 of the preprocessing unit 1 performs the first conversion in order to obtain a character pattern required for character recognition.
The multi-valued image stored in the pattern register 12 is binarized to obtain a character pattern, which is input to the second pattern register 31 (FIG. 3). In addition, 2 here
As described above, the binarization process is performed on the multivalued image with R,
RG of each pixel is set by setting an appropriate threshold for each of G and B.
If all the B values are less than or equal to the threshold value, the black pixel (1) is set, and otherwise, the white pixel (0) is set. However, the method of binarization processing is not limited to this and can be any method.

The character pattern is the second pattern register 31.
When stored in, the line width calculator 32 (FIG. 3) detects the line width of this character pattern. This is performed as follows in this embodiment. The line width calculation unit 32 has a shift register configuration like a well-known filter circuit, and for this character pattern, the total number Q of points where all 2 × 2 windows are black bits and the total number A of black bits are set. Count and calculate the line width w using the well known approximation formula (1) below.

W = 1 / {1- (Q / A)} (1) Further, the vertical sub-pattern extraction unit 34 performs a vertical scan on the second pattern register 31, and the result w of the line width calculation unit is obtained. Is used as a parameter, the vertical sub-pattern (hereinafter referred to as “V”) is deleted by deleting a black bit portion whose continuous length S of black bits in the scanning direction does not satisfy the following relational expression (2).
It may be abbreviated as "SP". ) Get.

S ≧ N · w (2) It should be noted that N in the equation (2) is a value determined according to the type of the writing instrument determined by the writing instrument determination unit 2. In this embodiment, each of the sub-pattern extraction units 34 to 37 is configured to select this N based on the information indicating the type of writing instrument input from the writing instrument determination unit 2. N to choose
Can be stored in a suitable table ROM and read out. Specifically, in the case of a writing instrument which tends to have many blurred characters, N is selected to a small value. Also, in the case of a writing instrument which tends to have many collapsed characters, N is set to a large value. Since N is changed according to the type of writing instrument as described above, it is possible to obtain a stable sub-pattern that absorbs deformation of characters such as blurring or blurring that occurs depending on the type of writing instrument used when writing characters on the medium to be read.

Similar to the operation of the vertical sub-pattern 34, the horizontal sub-pattern extraction unit 35 scans a horizontal sub-pattern (HSP) by horizontal scanning, and the right diagonal sub-pattern extraction unit 36 scans a right diagonal 45 ° by a right diagonal sub-pattern. The left diagonal sub-pattern extracting unit 37 extracts the left diagonal sub-pattern (LSP) by scanning the left diagonal 45 °. VSP, HSP, RSP and LSP extracted as above from the character pattern of katakana "e"
Together with the character pattern of "e" in FIG.
Each is shown in (E).

Further, the character frame detecting section 33 detects the area corresponding to the character frame circumscribing the character pattern in the second pattern register 31 and sends the result to the feature matrix extracting section 38. The detected character frame in this case is shown in FIG.
In (E), it becomes a square indicated by a broken line. The character frame can be detected, for example, by scanning the inside of the second pattern register 31 and detecting the boundary frame in which the black bit appears.

Further, the feature matrix extraction unit 38 uses the second
The region corresponding to the character frame in the pattern register 31 is divided into m × n regions for each of the vertical, horizontal, right diagonal, and left diagonal sub patterns. In FIG. 7, m = n = 5.
This is an example in which the vertical sub-pattern of "e" shown in FIG. Further, the feature matrix extraction unit 38 counts the number of black bits B _ij of each region thus divided, and uses the line width w to _{express the} feature L _ij indicating the line length by the following equation (3). Is calculated, and here, m × n × 4
Create a dimensional feature matrix. Where B _ij and L _ij
In the above, i is a symbol indicating a horizontal element in each component of the feature matrix, and j is a symbol indicating a vertical element.

L _ij = B _ij / w (3) It should be noted that the feature matrix thus obtained is generally subjected to normalization processing to obtain a final m × n × 4 normalized feature matrix ( That is, the feature amount of the character pattern) is obtained.
This normalization is typically performed for the VSP feature matrix with the y-axis length ΔY of the character frame, for the HSP feature matrix with the x-axis length ΔX, and for each of the RSP and LSP feature matrices. Is {(ΔX) ² +
This can be done by normalizing with (ΔY) ² } ^1/2 .

The discriminating unit 4 is arranged between the characteristic amount of a plurality of standard character patterns prepared in advance, so-called standard character mask (fm), and the characteristic matrix (fi) created by the characteristic matrix extracting unit 38. The distance D given by the well-known formula (4), that is, the category name of the standard mask that gives the smallest value of the difference vector length of two vectors in the m × n × 4 dimensional feature space is output as the character name output. .

D = {Σ (fm-fi) ² } ^1/2 (4) As is apparent from the above description, according to the first aspect of the invention, when the feature of the character pattern is extracted by the feature extracting unit. In addition, since the processing is performed in consideration of the determination result of the type of writing instrument, it is possible to extract a stable feature amount in which blurring or crushing caused by the type of writing instrument is removed. For this reason, even if the subsequent recognition processing is performed as usual, the reduction in the recognition rate due to the difference in the writing instrument is reduced.

2. Description of the Second Invention In the first invention described above, the feature extraction unit 3 corrects the blurring and the crushing due to the type of the writing instrument used when writing characters on the medium to be read. You can also do it. This second invention is an example.

FIG. 8 is a block diagram showing the configuration of the character recognition device according to the second embodiment of the present invention. The character recognition device according to the second embodiment of the present invention includes a preprocessing unit 1, a writing instrument determination unit 2, a feature extraction unit 3a, a blur processing unit 5, and an identification unit 4. Of these constituent components, the preprocessing unit 1, the writing instrument determination unit 2, and the identification unit 4 can be configured by the same units as the respective units of the character recognition device according to the embodiment of the first invention. However, the writing instrument determination unit 2 outputs the writing instrument determination result to the blur processing unit 5. Therefore, the configuration of the feature extraction unit 3a and the blurring processing unit 5 which are constituent components different from those of the character recognition device of the first invention will be described below.

First, the feature extraction unit 3a in the second invention.
9, as shown in FIG. 9, the second pattern register 31, the line width calculation unit 32, the character frame detection unit 33, the vertical sub-pattern extraction unit 34a, the horizontal sub-pattern extraction unit 35a, the right diagonal sub-pattern extraction unit 36a, Left diagonal sub-pattern extraction unit 3
7a and a characteristic matrix extraction unit 38.
Of these, the second pattern register 31, the line width calculation unit 32, the character frame detection unit 33, and the feature matrix extraction unit 3
8 has the same structure as each part in the first invention.
On the other hand, with respect to each of the sub-pattern extraction units 34a to 37a, in the first invention, the information indicating the type of the writing tool is input from the writing tool determination unit 2 to each sub-pattern extraction unit, and black bit extraction is controlled based on the information. On the other hand, this second invention does not do this. That is, each of the sub-pattern extraction units 34a to 37a deletes a black bit portion in which the continuous length S of black bits in each scanning direction does not satisfy the following relationship (5) with respect to the line width w, thereby removing each direction. Subpatterns VSP, HSP, RSP, LSP of However,
T is a predetermined constant.

S ≧ T · w (5) Therefore, each sub-pattern extracted in this case is
It will include features due to differences in writing instruments. This will be corrected in the blur processing unit later.

Further, the feature matrix extraction unit 38 (FIG. 9) of the feature extraction unit 3a creates a normalized feature matrix for each sub-pattern as in the case of the first invention.

The blur processing unit 5 is the writing instrument discrimination unit 2 described above.
The degree of blurring processing applied to the feature amount of the character pattern (in this case, the normalized feature matrix obtained by the feature extraction unit 3a) is varied according to the determination result obtained in (1). Therefore, the blur processing unit 5 in the case of this embodiment
Has a configuration that operates as follows.

That is, the blurring processing unit 5 performs blurring processing on the normalized feature matrix extracted by the feature extracting unit 3a in the second invention in the subpattern extraction direction by the following method in this embodiment. do.

First, for the elements M _ij of the vertical feature matrix created by the feature matrix extractor 38 based on the vertical sub-patterns extracted by the vertical sub-pattern extractor 34a, the following equation (6) is applied to blur. Compute the element M _{Bij of the} feature matrix. Here, i in M _ij and M _Bij is a symbol indicating a horizontal element in each component of the feature matrix, and j is a symbol indicating a vertical element. Further, C is a value determined according to the discrimination result of the writing instrument discriminating unit 2 (the same applies to the following equations (7) to (9)).
Specifically, when the number of faint characters tends to increase, C is reduced to increase the degree of blurring. When the number of characters that are blurred is likely to increase, C is increased to reduce the degree of blurring. When calculating the outermost area of the elements M _ij , the elements (M _{i + 1, j} or M _{i, j + 1} ) in the further outer area are regarded as equal to the outermost element ( The same applies to the following expressions (7) to (9). That is, each element of the feature matrix on the right side of each of the above equations is set to M
_p, the subscript p when you want to collectively referred to as _q, for q, p
> M or q> n (m and n are values that determine the above-mentioned divided areas) or p <1 or q <1 is satisfied, M _pq
= M _ij (same in the following expressions (7) to (9)).

[0037]

[Equation 1]

This equation (6) is used for the vertical matrix element value M _{ij of} interest and the element values M _{i-1, j} adjacent to it in the horizontal direction.
And M _{i + 1, j} are used to mean blurring with a weighted average value. Here, each element value is an amount associated with the length in each region of the vertical sub-pattern, so
After all, it can be considered that the expression (6) redistributes the length of each element in the horizontal direction.

Similarly, for the elements M _ij of the horizontal feature matrix created by the feature matrix extractor 38 based on the horizontal sub-patterns extracted by the horizontal sub-pattern extractor 35a, the following equation (7) is applied to blur. Compute the element M _{Bij of the} feature matrix. It can be considered that the expression (7) redistributes the length of each element of the horizontal feature matrix in the vertical direction.

[0040]

[Equation 2]

Similarly, the right diagonal sub-pattern extraction section 36a
The element M _Bij of the blur feature matrix is calculated by applying the following equation (8) to the element M _ij of the right diagonal feature matrix created by the feature matrix extracting unit 38 on the basis of the right diagonal sub-pattern extracted in. It can be considered that the equation (8) redistributes the length of each element of the right diagonal feature matrix in the left diagonal direction.

[0042]

(Equation 3)

Similarly, the left diagonal sub-pattern extracting section 37a
With respect to the element M _ij of the left diagonal feature matrix created by the feature matrix extracting unit 38 based on the left diagonal sub-pattern extracted in step (4), the following equation (9) is applied to calculate the element M _Bij of the blur feature matrix. It can be considered that the expression (9) redistributes the length of each element of the left diagonal feature matrix in the right diagonal direction.

[0044]

(Equation 4)

As described above, since the blur feature matrix element is changed according to the type of writing instrument with C as a parameter, deformation such as blurring or crushing that occurs depending on the type of writing instrument when a character pattern is written on a medium to be read is blurred. The process can be absorbed and thus a stable blur feature matrix is obtained. After that, the identification unit 4 compares the blur feature matrix with the feature amounts of a plurality of standard character patterns prepared in advance, and identifies the character pattern. This is basically the same as in the first aspect of the invention, so description thereof will be omitted.

Although the embodiments of the inventions of this application have been described above, the invention is not limited to the above-mentioned embodiments.

For example, in each of the above-described embodiments, an example in which the writing instrument is a pencil or a ballpoint pen has been described, but the present invention can be applied to any other writing instrument, such as a felt-tip pen. Further, in the above-described embodiment, the extraction of the pixels forming the character from the input image and the determination of the writing tool are both performed based on the color information of R, G, B of the pixel. Can also perform the above-mentioned pixel extraction and writing instrument discrimination by paying attention to the difference in brightness. Of course, the present invention can be applied to the case where the medium to be read is not a color image but a monochrome image (monochrome image data).

[0048]

As is apparent from the above description, according to the character recognition device of the first invention of the present application, in the character recognition device having the writing instrument discrimination part, the feature extraction part and the identification part, the feature extraction is performed. Since the part is configured to have a different feature amount to be extracted according to the determination result obtained by the writing instrument determination unit, it is possible to extract a feature amount that can remove blurring or crushing due to the difference in the type of writing instrument. In addition, recognition processing can be performed using this feature amount. Further, according to the character recognition device of the second invention, in a character recognition device comprising a writing tool discrimination unit, a feature extraction unit, a blur processing unit and an identification unit,
Since the blur processing unit is configured to change the degree of blur processing performed on the feature amount of the character pattern according to the determination result obtained by the writing instrument determination unit, blurring caused by a difference in writing instrument type or Blurring processing capable of removing crushing and the like can be performed, and recognition processing can be performed using the blur feature matrix obtained thereby. Therefore, in these first and second inventions, it is possible to correctly read the character pattern that has been an error or rejected due to the difference in the type of writing instrument used when writing characters or the like on the medium to be read. A higher character recognition device can be realized.

[Brief description of drawings]

FIG. 1 is an explanatory diagram of a character recognition device according to an embodiment of the first invention.

FIG. 2 is an explanatory diagram of a configuration example of a preprocessing unit in the character recognition device of the first and second embodiments.

FIG. 3 is an explanatory diagram of a configuration example of a feature extraction unit in the character recognition device according to the embodiment of the first invention.

FIG. 4 is an explanatory diagram of a medium to be read according to the embodiment.

FIG. 5 is an explanatory diagram of processing in a pattern register.

FIG. 6 is an explanatory diagram of an example of extracting a sub pattern.

FIG. 7 is an explanatory diagram of an example of dividing a character frame.

FIG. 8 is an explanatory diagram of a configuration example of a character recognition device according to an embodiment of the second invention.

FIG. 9 is an explanatory diagram of a configuration example of a feature extraction unit in the character recognition device according to the second embodiment of the invention.

[Explanation of symbols]

 1: Preprocessing part 2: Writing instrument discrimination part 3: Feature extraction part 3a: Feature extraction part in the second invention 4: Identification part 5: Blur processing part

Claims (3)

[Claims] 1. A writing instrument discriminating unit for discriminating the type of writing instrument used when a character is written on a medium to be read based on image data of the medium to be read, and a preprocessing unit for obtaining a character pattern from the image data. A feature extraction unit for extracting the feature amount of the character pattern, and an identification for identifying the character pattern by collating the extracted feature amount of the character pattern with the feature amounts of a plurality of standard character patterns prepared in advance. A character recognition device including a section, the character extraction device being configured to change the feature amount to be extracted by the feature extraction unit according to a determination result obtained by the writing instrument determination unit. 2. The character recognition device according to claim 1, wherein the feature extraction unit includes a line width calculation unit that detects a line width w from the character pattern, and the character pattern is scanned in a plurality of directions. A group of black bits satisfying S ≧ N · w (N is a value determined according to the discrimination result in the writing instrument discriminating section) between the continuous length S of black bits in a plurality of scanning directions and the line width w. By extracting a plurality of sub-patterns representing stroke components for each of the plurality of scanning-direction components, and m × n (m, n is 1 for each sub-pattern extracted by the sub-pattern extraction unit. The above positive integers may be the same as each other.), And the feature quantity is calculated using the number of black bits and the line width in each partial area for each sub-pattern. Features as A character recognition device comprising a feature matrix extraction unit for creating a lix. 3. A writing instrument discriminating unit which discriminates the type of writing instrument used when a character is written on a medium to be read based on image data of the medium to be read, and a preprocessing unit which obtains a character pattern from the image data. , A feature extraction unit that extracts the feature amount of the character pattern, a blur processing unit that performs blur processing on the extracted feature amount of the character pattern, and a feature amount that has been subjected to the blur processing of the character pattern are prepared in advance. In a character recognition device comprising an identification unit that identifies the character pattern by comparing it with the characteristic amount of a plurality of standard character patterns, the blurring processing unit, in the determination result obtained by the writing instrument determination unit. The character recognition device is configured to change the degree of blurring processing applied to the characteristic amount of the character pattern accordingly.