JPH10187886A - Character recognizing device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To normalize the character line width at almost the fixed value and to accurately recognize the characters despite the large variance of the character line width by calculating the feature vector from the border line direction data having the normalized line width and recognizing the character based on the feature vector. SOLUTION: At a character recognition part 5, the border line direction is detected at each position of a character image and the border line direction data are acquired. Then the border line data are equality divided into seven pieces in both vertical and horizontal directions, for example, so that the images divided into (7×7) blocks are obtained. The feature vector is calculated by counting the number of direction data included in every block of the divided images. Then the feature vector is compared with a reference pattern (standard pattern) that is previously produced and stored in a reference pattern file 6 for the execution of recognition of characters. The obtained results of recognition are outputted and stored in a recognition result file 7.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognizing apparatus and a character recognizing method for recognizing a character by using a distribution of the direction of the outline of the character as a feature.

[0002]

2. Description of the Related Art Conventionally, in a character recognition apparatus for recognizing a character on an object to be read, a method of performing character recognition using a distribution of a direction of a contour of the character as a feature amount has been widely used. As an example, see, for example,
There is a technique disclosed in Japanese Patent No. 3793.

[0003] Briefly explaining this conventional technique, by applying a 2x2 logical filter to an input character image, the direction of the outline at each position of the character image is detected to obtain outline direction data. The obtained contour direction data is equally divided vertically and horizontally into seven to obtain an image divided into a plurality of blocks, and the number of direction data in each of the obtained blocks is counted to obtain a feature vector. The obtained feature vector is compared with a reference pattern created in advance, and character recognition is performed by the composite similarity method. Such features of the directional distribution of the outline of a character are relatively widely used in recent character recognition devices because they are relatively stable against deformation of the character.

[0004]

In a conventional character recognition apparatus, only a ballpoint pen, a pencil and a mechanical pencil have been used as writing tools for handwritten characters. , Brush pen,
It has become necessary to read handwritten characters written with magic ink or the like.

[0005] Also, in the recognition of print characters, it has conventionally been necessary to recognize only a limited number of characters, but in recent years it has become necessary to recognize very various characters. As a result, the character line widths of the characters to be read have become very diverse from thin to thick.

However, the conventional character recognition method which is characterized by the directional distribution of the outline of a character cannot sufficiently cope with such a large variation in line width. For example, FIG.
If a thick character such as 0 is input,
The position of the outline is greatly shifted as compared with the character having the normal line width shown in FIG. 3, and the outline is often moved to a different block. As a result, the value of the feature vector also fluctuates greatly, and there is a high possibility that an incorrect character recognition result will be obtained.

Therefore, according to the present invention, when recognizing a character by using the distribution of the direction of the outline of the character as a feature, the character line width can be normalized to a substantially constant value by a simple method. It is an object of the present invention to provide a character recognition device and a character recognition method capable of performing accurate character recognition even when the variation of the character is large.

[0008]

A character recognition apparatus according to the present invention detects a contour direction at each position of an input character image in a character recognition apparatus for recognizing a character by using a distribution of a direction of a contour of a character as a feature quantity. Contour direction detecting means for obtaining contour line direction data, character line width calculating means for determining a character line width from an input character image, and an amount corresponding to the character line width determined by the character line width calculating means. Character line width normalizing means for normalizing the character line width to a constant value by moving the outline direction data obtained from the outline direction detecting means in a predetermined direction, and character line width by the character line width normalizing means. Are provided with feature vector calculating means for obtaining a feature vector from normalized contour direction data, and recognition means for performing character recognition using the feature vector obtained by the feature vector calculating means.

Further, the character recognition device of the present invention has input means for inputting an image on an object to be read, and line extraction for extracting an image of a character line on the object to be read from the image input by the input means. Means, character extracting means for extracting an individual character image from the image of the character line extracted by the line extracting means, and applying a predetermined logical filter to the character image extracted by the character extracting means. A contour direction detecting means for detecting contour direction at each position of the image to obtain contour direction data, and measuring a total number of black pixels and a total length of the contour of the character image extracted by the character extracting means, respectively. A character line width calculating means for obtaining a character line width using the total number of measured black pixels and the total length of the contour, a character line width obtained by the character line width calculating means, a predetermined normalized character line width, To ring The amount of movement of the line is obtained, and the contour line direction data obtained from the contour line direction detecting means is moved in the direction orthogonal to the direction data by the determined amount of movement of the contour line, so that the character line width is fixed. Character line width normalizing means for normalizing the contour line direction data having the character line width normalized by the character line width normalizing means into a plurality of blocks, and dividing the direction data in each of the divided blocks. The image processing apparatus includes a feature vector calculating unit that calculates a feature vector by counting the number, and a recognition unit that performs character recognition using the feature vector obtained by the feature vector calculating unit.

Further, according to the character recognition method of the present invention, in the character recognition method for character recognition using the distribution of the direction of the outline of the character as a feature amount, the outline direction at each position of the input character image is detected. Get the data,
The total number of black pixels and the total length of the outline of the input character image are measured, and the character line width of the character image is determined using the measured total number of black pixels and the total length of the outline, and the amount corresponding to the determined character line width. Only, the obtained contour direction data is
By moving in the direction orthogonal to the direction data, the character line width is normalized to a constant value, the contour line direction data in which the character line width is normalized is divided into a plurality of blocks, and each of the divided blocks is A feature vector is obtained by counting the number of direction data, and character recognition is performed using the obtained feature vector.

According to the present invention, the contour direction data at each position of the character image is calculated in the direction orthogonal to the direction data by an amount corresponding to the estimated character line width (character line width obtained from the input character image). By normalizing the character line width by moving the character line width, the character line width is normalized to a substantially constant value by a simple method when the character is recognized using the distribution of the direction of the outline of the character as a feature amount. This enables accurate character recognition even when the character line width varies greatly.

[0012]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 schematically shows a configuration of a character recognition device according to the present embodiment. This character recognition device includes an image input unit 1 including a photoelectric converter or the like as input means for inputting an image on a form P as an object to be read, and a memory for temporarily storing an image input by the image input unit 1. The image file 2 as a means, a line extracting unit 3 as a line extracting means for extracting an image of a character line on the form P from the image in the image file 2, and individual lines from the character line image extracted by the line extracting unit 3 A character extracting unit 4 as a character extracting unit for extracting a character image; a character recognizing unit 5 for performing character recognition from the character image extracted by the character extracting unit 4;
And a reference pattern file 6 as storage means for storing a reference pattern used for character recognition, and a recognition result file 7 for storing recognition results of the character recognition unit 5.

In such a configuration, the overall operation will be briefly described. First, an image on the form P is input by the image input unit 1 and stored in the image file 2. Next, the line extracting unit 3 performs predetermined image processing on the image in the image file 2 to extract the image of the character line entered on the form P. Next, the character extracting unit 4 extracts individual characters from the image of the extracted character line. Next, the character recognition unit 5 performs character recognition from the extracted individual character images using the composite similarity method, and outputs the recognition result to a recognition result file 7 such as a magnetic disk.

Hereinafter, the contents of the recognition processing in the character recognition unit 5 will be described in detail with reference to the flowchart shown in FIG. First, when a character image as shown in FIG. 3 is input (S1), for example,
For example, by applying a 2 × 2 logical filter as shown in FIG. 4, the contour direction at each position of the character image is detected, and contour direction data is obtained (S2). That is, it is considered that a vertical contour exists at a position corresponding to the logical filter in FIG. Similarly, FIG.
(B), (c), and (d) indicate upward and downward, respectively,
The outline in the left-right direction is shown. By performing such a logical filtering operation on the entire surface of the input image, the contour direction at each position of the character image can be detected, and contour direction data as shown in FIG. 5 is obtained. In this figure, the outline direction of each position is indicated by an arrow.

Next, the total number of black pixels and the total length of the outline of the input character image are measured (S3, S4). Next, using the measured total number of black pixels and the total length of the contour, an average character line width is calculated by the following equation (S5).

(Average character line width) = 2 × (total number of black pixels)
/ (Total length of contour) Next, using the average character line width obtained as described above,
The contour movement amount is calculated by the following equation (S6). Here, the normalized character line width is a predetermined constant, and the character line width is set to this value by normalization processing.

(Contour line moving amount) = ｛(average character line width) −
(Normalized character line width)｝ / 2 In this case, if the line width of the input character image is large, the outline movement amount is a positive value, and if the line width is small, the outline value is a negative value.

Next, the outline direction data at each position of the character image obtained in step S2 is moved by the outline movement amount obtained in step S6 (S7). in this case,
The moving direction is a direction orthogonal to the direction data. That is, each direction data is moved in the direction indicated by the white arrow shown in FIG. However, the direction shown in FIG. 6 is the moving direction when the contour line moving amount is positive, and moves in the opposite direction by 180 degrees when the contour line moving amount is negative.

As described above, by moving each contour line direction data, the contour line direction data is converted as shown in FIG. 7 (a) to FIG. 7 (b). As is clear from FIG. 7, the thick character line width of the input character image is reduced, and the input character image is converted into contour direction data having a predetermined character line width (normalized character line width). When the line width of the input character image is small, contour line direction data thickened to a certain character line width can be obtained by the same processing.

Next, FIG. 7B obtained in step S7
By dividing the contour direction data into 7 in each of the vertical and horizontal directions, an image divided into 7 × 7 blocks as shown in FIG. 8 is obtained (S8).

Next, as shown in FIG. 9, the number of each direction included in each block is arranged to form a feature vector (S
9). That is, the feature vector is calculated by counting the number of each direction data included in each block of the image as shown in FIG. 8 obtained in step S8. In the case of this example, since the number of blocks in the four directions is 7 × 7,
The dimension number of the feature vector is 4 × 7 × 7 = 196 dimensions.

Next, character recognition is performed by comparing the feature vector obtained in step S9 with a reference pattern (standard pattern) previously created and stored in the reference pattern file 6 (S10). As a method of character recognition, for example, a composite similarity method may be used. Note that the reference pattern itself used here is also created from a character image whose character line width has been normalized by the same method.

Next, the recognition result obtained in step S10 is output and stored in the recognition result file 7 (S11).
As described above, according to the above embodiment, the contour direction data at each position of the character image is orthogonal to the direction data by an amount corresponding to the estimated character line width (character line width obtained from the input character image). By performing normalization of the character line width by moving in the direction in which the character line is moved, the width of the character line can be set to a substantially constant value by a simple method when recognizing the character using the distribution of the direction of the outline of the character as a feature amount. The character can be accurately recognized even when the character line width greatly fluctuates, for example, when a character image having a large character line width as shown in FIG. 10 is input.

The present invention is not limited to the above-described embodiment, but can be variously modified without departing from the spirit of the present invention. For example, in the above embodiment, the overall average character line width is used as the character line width of the input character image. However, there are often cases in which the thickness of character lines constituting characters is not uniform. In such a case, if line width normalization is performed using the entire average character line width, the character image after the normalization may be distorted.

In such a case, a local character line width may be obtained for each position of the character image, and a contour line moving amount may be locally calculated in accordance with the local character line width. The local character line width can be obtained by measuring the distance from a certain contour position to a contour position existing in the opposite direction, as shown in FIG.
However, as shown in FIG. 12, when the distance is very different from the global average character line width, it is determined that the character line width does not correspond to the correct character line width, and the contour line direction data is not moved. .

For example, when the normalized character line width is "1"
, A thinned image as shown in FIGS. 13A to 13B is obtained by the normalization processing. Such a thinned image can be used for various purposes such as detection of an end point or an intersection of a character line.

[0027]

As described above in detail, according to the present invention, when recognizing a character using the distribution of the direction of the contour of the character as a feature, the character line width is normalized to a substantially constant value by a simple method. Thus, a character recognition device and a character recognition method capable of performing accurate character recognition even when the variation of the character line width is large can be provided.

[Brief description of the drawings]

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

FIG. 2 is a flowchart illustrating a character recognition process.

FIG. 3 is a diagram illustrating an example of an input character image.

FIG. 4 is a diagram showing a logical filter used for detecting the direction of the outline of a character.

FIG. 5 is a diagram illustrating a distribution of a direction of a contour line of a character.

FIG. 6 is a diagram showing a moving direction of each contour line direction data.

FIG. 7 is a diagram showing that character line width normalization is performed by moving contour line direction data.

FIG. 8 is a diagram illustrating an image obtained by dividing contour line direction data of a character image into a plurality of blocks.

FIG. 9 is a diagram illustrating an example of calculating a feature vector.

FIG. 10 is a diagram illustrating a problem when a character image having a thick line width is input.

FIG. 11 is a diagram illustrating a method for calculating a local character line width.

FIG. 12 is a view for explaining correspondence that is not regarded as a local character line width.

FIG. 13 is a view for explaining that a thinned image can also be obtained by normalization processing.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Image input part (image input means) 2 ... Image file (storage means) 3 ... Line extraction part (line extraction means) 4 ... Character extraction part (character extraction means) 5 ... Character recognition part 6 ... ... Reference pattern file (storage means) 7 ... Recognition result file (storage means)

Claims (11)

[Claims] 1. A character recognition apparatus for recognizing a character based on the distribution of the direction of the outline of the character as a feature amount. A contour direction detecting means for detecting a contour direction at each position of an input character image to obtain contour direction data. Character line width calculating means for calculating a character line width from an input character image; and contour line direction data obtained from the contour line direction detecting means by an amount corresponding to the character line width calculated by the character line width calculating means. Character line width normalizing means for normalizing the character line width to a fixed value by moving the character line width in a predetermined direction, and a feature vector from the contour direction data in which the character line width is normalized by the character line width normalizing means. And a recognition unit for performing character recognition using the feature vector obtained by the feature vector calculation unit. 2. The contour line direction detecting means detects a contour line direction at each position of a character image by applying a 2 × 2 logical filter to the input character image. Character recognition device. 3. The character line width calculating means measures the total number of black pixels and the total length of the outline of the character image, respectively, and calculates the character line width of the character image using the measured total number of black pixels and the total length of the outline. The character recognition device according to claim 1, wherein the character recognition is performed. 4. The character line width calculating unit according to claim 1, wherein an average character line width of the entire character image is obtained.
Or the character recognition device according to 3. 5. The character recognition device according to claim 1, wherein said character line width calculating means obtains a local character line width at each position of the character image. 6. The character line width calculating unit calculates a global character line width of a character image and a local character line width at each position of the character image.
Or the character recognition device according to 3. 7. The character line width normalizing means, wherein the character line width calculated by the character line width calculating means and a predetermined normalized character line width are used to determine a contour moving amount. Contour calculating means for moving the contour direction data obtained from the contour direction detecting means in a direction orthogonal to the direction data by the amount of movement of the contour determined by the contour moving amount calculating means. 2. The character recognition device according to claim 1, further comprising a moving unit. 8. The feature vector calculating means divides the contour line direction data whose character line width has been normalized by the character line width normalizing means into a plurality of blocks, and calculates the number of direction data in each of the divided blocks. 2. The character recognition device according to claim 1, wherein a feature vector is obtained by counting the number of characters. 9. The method according to claim 1, wherein the recognition unit compares the feature vector obtained by the feature vector calculation unit with a reference pattern created in advance and performs character recognition by a composite similarity method. Character recognition device according to the description. 10. An input means for inputting an image on an object to be read, a line extracting means for extracting an image of a character line on the object to be read from the image input by the input means, A character extracting means for extracting an individual character image from the image of the extracted character line; and a contour line direction at each position of the character image by applying a predetermined logical filter to the character image extracted by the character extracting means. And the contour line direction detecting means for detecting contour line direction data, and measuring the total number of black pixels and the total length of the contour of the character image extracted by the character extracting means, respectively. Character line width calculating means for calculating the character line width using the entire length of the character line width; and determining the moving amount of the contour line from the character line width calculated by the character line width calculating means and a predetermined normalized character line width. ,this The character line width normalization is performed by normalizing the character line width to a constant value by moving the outline direction data obtained from the outline direction detecting means by the amount of movement of the outline in the direction orthogonal to the direction data. The contour line direction data whose character line width has been normalized by the character line width normalizing means is divided into a plurality of blocks, and the number of direction data in each of the divided blocks is counted. And a recognition unit for performing character recognition using the feature vector obtained by the feature vector calculation unit. 11. A character recognition method for recognizing a character using a distribution of the direction of the outline of the character as a feature quantity, wherein the outline direction at each position of the input character image is detected to obtain outline direction data. The total number of black pixels of the image and the total length of the outline are measured, and the character line width of the character image is obtained using the measured total number of black pixels and the entire length of the outline. The obtained contour line direction data is moved in a direction orthogonal to the direction data to normalize the character line width to a constant value, and the character line width normalized contour line direction data is divided into a plurality of blocks. A character is obtained by dividing and calculating a feature vector by counting the number of direction data in each of the divided blocks, and performing character recognition using the obtained feature vector.識方 method.