JP2001291056A - Document picture recognizing device and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To properly and quickly binarize a character picture and the other picture area whose characteristics are different from an inputted multi-level picture. SOLUTION: In this document picture recognizing device equipped with a picture binarizing means 10a for preparing a binary picture from an input multi- level picture 1 and a picture recognizing means 14a for recognizing the prepared binary picture, the picture binarizing means 10a separates an input multi-level picture 1 into a character picture area and a background picture area, and binarizes each separated character picture area, and decides the binarization threshold of the background picture area from the binarization threshold at the time of the binarization processing, and binarizes the background picture area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a document image recognition apparatus and a recording medium for recognizing characters from a multi-valued image captured by a scanner, a digital camera, or the like.

[0002]

2. Description of the Related Art When analyzing features of an image, a binary image in which a target graphic and a background are separated from a multi-valued image is often created and handled. The binarization of the image is performed by a threshold value process in which when the gradation value of the multi-valued image is equal to or smaller than the threshold value, the image is set to black, and when it exceeds the threshold value, it is set to white. Usually, the black part of the binary image represents the target figure, and the white part represents the background.

[0003] As a method of determining a threshold value, when a distribution having two peaks is obtained as a result of obtaining a gradation histogram of a given multi-valued image, a valley between these two peaks is defined as a threshold value. Assuming that a set of tone values is divided into two classes (more than t and less than t) by a threshold value t in a mode method or a tone histogram, the separation (variance value) between the two classes is Determine the parameter (threshold value t) to be the best (literature: IEICE Transactions on Electronics 80/4
Vol. J63-D No. 4, p. 349-356), and binarization processing using a discriminant criterion in discriminant analysis.

In a document image recognition apparatus, a binary image created by such a binarization process is treated as a recognition image, and characters are extracted (coded).

[0005]

SUMMARY OF THE INVENTION
There were the following issues.

In the conventional binarization process, if the background of a character image to be extracted has a color, the density of the read character image and the background color are close to each other. It became difficult, and a part of the background area was sometimes represented as a black pixel together with the character image in the created binary image. When character recognition is performed using such a binary image, there has been a problem that recognition accuracy is reduced. In addition, such a binary image has a problem that it is not suitable as an image for display from the point of viewability.

An object of the present invention is to solve such a conventional problem and appropriately and rapidly binarize a character image and other image regions having different properties from an input multi-valued image. I do.

[0008]

FIG. 1 is a diagram illustrating the principle of the present invention. In FIG. 1, 1 is an input multi-value image, 10a is an image 2
The value conversion means 14a is an image recognition means.

The present invention is configured as follows in order to solve the above-mentioned conventional problems.

(1): A document image recognition apparatus provided with an image binarizing means 10a for creating a binary image from an input multi-valued image 1 and an image recognizing means 14a for recognizing the created binary image. The image binarizing means 10a separates the input multi-valued image 1 into a character image area and a background image area,
Performing a binarization process on each of the separated character image regions;
The binarization threshold of the background image area is determined from the binarization threshold at the time of the binarization processing, and the binarization processing of the background image area is performed.

(2): In the document image recognition device of (1), the image binarizing means 10a selects a low gradation representative value included in the input multi-valued image 1 and performs a binarizing process. Then, the generated binary image is separated into a character image region including many character images and a background image region other than the character image region.

(3) In the document image recognition apparatus of (1), the image binarizing means 10a is configured to determine whether the separated character image area has a single gradation when the entire character image area has a single gradation. The character image area is changed so that the background image is included in the gradation character image area, and the binarization process is performed.

(4) In the document image recognition apparatus of (1), the image binarizing means 10a performs each binarization threshold in the binarization processing performed on each of the separated character image areas. The area within the rectangle enlarged to include the binarized area is calculated by the obtained binarization threshold value.
Perform value processing.

(5): The input multi-valued image is separated into a character image region and a background image region, and a binarization process is performed on each of the separated character image regions. Image binarizing means 10 for determining a binarization threshold value of the background image region from a binarization threshold value and performing binarization processing of the background image region
a, and an image recognizing unit 14a for recognizing the binary image created by the binarization process, and a computer-readable recording medium storing a program for causing a computer to function.

(Operation) The operation based on the above configuration will be described.

A binary image is created from the input multivalued image 1 by the image binarizing means 10a, and the binary image created by the image recognizing means 14a.
In a document image recognition apparatus for recognizing a value image, the input multi-valued image 1 is separated into a character image region and a background image region by the image binarizing means 10a. A binarization process is performed, and a binarization threshold of the background image region is determined from the binarization threshold value at the time of the binarization process, and the binarization process of the background image region is performed. For this reason, even if the image of the character image has a color on the background, the image can be binarized with high accuracy, and the binarization can be performed at a high speed because the binarization process is performed with a fixed threshold value.

Further, the image binarizing means 10a selects a low gradation representative value included in the input multi-valued image 1 and performs a binarizing process, and converts a character image from the created binary image. It is separated into a character image area including many and a background image area other than that. For this reason, it is possible to binarize only a black portion having a low gradation and easily separate a character image region.

Further, in the image binarizing means 10a, when the separated character image area has a single gradation of the character image, the background image is included in the single gradation character image area. Then, the character image area is changed and the binarization process is performed. For this reason, even when the entire character image area is a character image, it is possible to prevent the character image from being output thinly,
Value conversion can be performed.

The image binarizing means 10a obtains each binarization threshold value at the time of the binarization process performed on each of the separated character image regions, and determines the binarized region. The area within the rectangle enlarged so as to be included is subjected to binarization processing using the obtained binarization threshold. For this reason, a character image portion that has not been extracted can be appropriately binarized as a character image region.

Further, the input multi-valued image is separated into a character image region and a background image region, and each of the separated character image regions is subjected to a binarization process, and a binarization threshold in the binarization process. Image binarization means 10 for determining a binarization threshold value of the background image area from the value and performing a binarization process of the background image area
a, and an image recognizing unit 14a for recognizing the binary image created by the binarization process, and a computer-readable recording medium storing a program for causing a computer to function. Therefore, by installing the program on the recording medium into a computer, a document that can be binarized with high precision and at high speed even if the image has a character image with a colored background. An image recognition device can be easily provided.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A document image recognition apparatus according to the present invention separates an input multi-valued image into an image area containing many character images and an image area other than the image area. It is provided with a binarization processing means for performing different image processing and creating a binary image at high speed.

Also, a low gradation (black) representative value included in the character image of the input multi-valued image is selected and binarized, and pixels included in the character image are extracted from the generated binary image. And a binary image area separating means for separating an image area containing many character images from other background image areas.

Further, a binarization process using a discrimination criterion in discriminant analysis is individually performed on an image region containing a large number of character images obtained by the binary image region separating means,
From the character image area binarizing means for creating a binary image and the variance of the character image area obtained by the character image area binarizing means, it is determined that the inside of the character image area has a single gradation. A single tone image area determining means, and a character image area determined to be a single tone by the single tone image area determining means,
A single-tone image area binarizing means for expanding a character image area so as to include a background image and performing binarization processing again using a criterion in the discriminant analysis.

Further, an optimum threshold value is obtained from the threshold values of the individual character image areas obtained by the character image area binarizing means, and the binarized area is obtained by the character image area binarizing means. The background image area is binarized by binarizing the background image area in the n × m rectangle including the character image area, thereby performing a binarization process on a character image portion not included in the character image area.

By providing such means, a binary image separated from a character image and an image region having other properties from an input multi-valued image is created, which is a problem in the prior art. Suitable binarization of a character image having a background color can be performed.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(1) Description of the Configuration of the Document Image Recognition Apparatus FIG. 2 is an explanatory diagram of the configuration of the document image recognition apparatus according to the embodiment of the present invention. In FIG. 2, the document image recognition apparatus includes a multi-value image binarization unit 10 and a binary image recognition unit 14. The multi-value image binarizing unit 10 includes a character image region detecting unit 11, a character image region binarizing unit 12, and a background image region binarizing unit 13.

This document image recognition device receives a multi-valued image (gray scale image) as an input,
At 0, each pixel is converted into a binary image having a value meaning either the character image area or the background image area.

The character image area detecting section 11 executes a binarization process on the entire gray scale image with a typical threshold value to determine an area containing a large number of character images. The character image area binarizing unit 12 individually performs a binarization process on an area including many character images. The background image area binarization unit 13 performs a binarization process on the background image area based on the threshold value of the character image area, and outputs a final binary image. The binary image recognizing unit 14 recognizes a character from the binary image created by the multi-value image binarizing unit 10,
It outputs the recognition result.

(2) Description of Character Image Area Detecting Section of Multivalued Image Binarizing Section FIG. 3 is an explanatory diagram of processing of the character image area detecting section. FIG.
In the processing of the character image area detecting section 11, first, a constant threshold value binarization processing S1 is performed, and then a labeling processing S1 is performed.
2, and finally, an effective segment area extraction process (character image area detection) S3 is performed (binary image area separation means).

In the constant threshold binarization process S1, binarization is performed on the input gray scale image with a constant threshold. As the threshold value at this time, a gradation is selected at which only the darker part in the character image is binarized.

In the labeling process S2, a set of connected components of the binary image obtained in the constant threshold binarization process S1 is extracted by a labeling process.

In the effective segment area extraction processing (character image area detection) S3, an effective segment is selected by eliminating connected portions of a size unsuitable for a character image (excluding ruled lines and 1-dot dust depending on the size). Then, a circumscribed rectangle of the selected segment is taken as a character image area.

FIG. 4 is an explanatory diagram based on an image of the binary image area separating means.
(A) is an input grayscale image. FIG.
In (a), the character image is composed of light black parts (gray parts) a1, b1 and darker parts a2, b2,
The background is the white part.

FIG. 4B shows a constant threshold value binarization process S1.
Is the processing result. In FIG. 4B, in the process of the constant threshold value binarization process S1, only the darker portions a2 and b2 in the character image are binarized by the constant threshold value, and the labeling process S2 is performed. Then, the character image area detection S3 is performed to extract a circumscribed rectangle (shown by a dotted line) of the darker portions a2 and b2.

FIG. 4C illustrates a gray scale image separated into a background area and a character image area. FIG.
In (c), the circumscribed rectangular coordinates of FIG. 4 (b) are adapted to a grayscale image (see FIG. 4 (a)), and a character image area including a background area (within a dotted line) and another background area (Outside the dotted line).

The reason why the circumscribed rectangle is obtained in this way is to include the character image area and the background image area other than the blacker part in the character image in the partial binarized area. In the binarization processing using the discriminant criterion in the discriminant analysis used in the binarization processing of the present embodiment, an area including an extraction target gradation image (character image) and a non-extraction target image (background image) By doing so, the optimal binarization is performed.

As a conventional method for determining a character image area, a Sobel filter process for extracting edges of the entire character image (calculating a change in gradation between pixels) may be used. On the other hand, in the present invention, the amount of calculation is reduced by using the constant threshold value binarization processing, and the character image area is detected at higher speed.

(3): Explanation of the character image area binarization section of the multi-value image binarization section In the character image area binarization section 12, the character image area detection section 1
The binarization process is performed on a region including a large number of character images determined in step 1 (character image region of a circumscribed rectangle) using a discrimination criterion in discriminant analysis for each region.

FIG. 5 is an explanatory diagram of the character image area binarizing section, and FIG. 5A is an explanatory diagram of the processing of the character image area binarizing section. In FIG. 5A, the character image area binarization unit 12
First, the binarization threshold value / variance value calculation processing S11 of the character image area is performed, and then the determination S12 is performed based on the variance value in the area to determine whether the background is included in the character image area. In this determination, if the background is included in the character image area, the binarization processing S13 in the character image area is performed. If the background is not included, the character image area expansion processing S14 is performed and the character image area is re-executed. Thresholding / variance calculating process S
Return to 11.

In the character image area binarization threshold value / variance calculation processing S11, first, for each character image area, the binarization threshold value is calculated using the criterion in the discriminant analysis, and the in-class variance value is calculated. Calculation processing is performed.

(Description of Calculation Method of Binarization Threshold and Intra-Class Variance Value) Hereinafter, a method of calculating the binarization threshold value and the in-class variance value will be described. FIG. 5B is an explanation of the gradation histogram. In FIG. 5B, the horizontal axis represents the gradation, and the vertical axis represents the number of pixels. Note that the gradation changes from black to white from left to right.

Assuming that the threshold value of a given region is t, t
Pixels having the above-mentioned gradation and pixels having smaller values are divided into two groups. These two groups are referred to as class 1 and class 2. The number of pixels of class 1 is ω ₁ (t), the average gradation of class 1 is M ₁ (t), and the number of pixels of class 2 is ω
₂ (t), the average gradation of class 2 is M ₂ (t), and the average gradation of all pixels is M _T , the inter-class variance σB ² is given by the following equation.

ΣB ² = ω ₁ (M ₁ −M _T ) ² + ω ₂ (M ₂ −M
_T ) ² = ω ₁ ω ₂ (M ₁ −M ₂ ) ² Here, the value of t that maximizes the inter-class variance σB ² by changing t is determined, and is set as a binarization threshold value in the region. .

(Explanation of Judgment of Whether Background is Included in Character Image Area) Next, judgment S12 is performed based on the variance value within the area to determine whether the background is included in the character image area (single gradation image area determination). means). This is because when the entire character image area does not include a background image as a character image, the above-described method of calculating the binarization threshold using the criterion in the discriminant analysis senses a small gradation difference. Since the value is obtained, a phenomenon occurs in which the character image cannot be correctly binarized at the threshold value and the character image is output thinly.

FIG. 6 is a diagram illustrating an example of a binarization process using a discriminant criterion in discriminant analysis. FIG. 6A is a diagram illustrating an example in which the entire area is a character image and does not include a background image. In FIG. 6A, the entire character image area is a character image, and is composed of pixels of a1 gradation and pixels of a2 gradation, and all the pixels in the character image area are binarization target gradations. ing. In this case, since the binarization threshold value is obtained in the area, the character image is output thinly (only the pixels of the gradation of a2 are output) as shown in the right diagram of FIG. Become. For this reason, 2 using the discriminant criterion in discriminant analysis
In the binarization processing, binarization cannot be correctly performed.

When the entire character image area is a character image and does not include a background image, it can be determined from the fact that the inter-class variance of such an area takes a smaller value than that of the area including the background. For this reason, the inter-class variance threshold value is determined, and for an area equal to or less than the threshold value, area expansion is performed to capture the surrounding background (character image area expansion processing S1).
4) The binarization threshold and the inter-class variance are calculated again for the expanded new area.

FIG. 6B is an illustration of the single-tone image area binarizing means. In FIG. 6B, as shown on the left side of FIG. 6B, all the pixels in the character image area (the pixel of the gray scale of a1 and the pixel of the gray scale of a2) correspond to the gray scale to be binarized. If so, the background image (white portion) is captured by expanding (expanding) the area as shown in the center of FIG. 6B. next,
For the expanded area, the binarization threshold and the inter-class variance are calculated again using the criterion in the discriminant analysis.

For an area where the inter-class variance is larger than the inter-class variance threshold and the background image is determined to be included in the character image area, the obtained binarization threshold
A value image is created (character image area binarization means).

FIG. 6C is an explanation of an example of an area determined to include a character image and a background image in the area. FIG.
In (c), if the background (white portion) is included in the character image area as shown in the left diagram, an appropriate threshold value is obtained by binarization processing using the discrimination standard in discriminant analysis. An appropriate binarization process is performed as shown in FIG.

By performing the above processing on all the character image areas determined by the character image area detection unit 11, the character image is binarized.

(4) Description of Background Image Area Binarization Unit In (3), the area detected by the character image area detection unit 11 was subjected to binarization processing. However, since this area is a circumscribed rectangle of a darker portion in the character image, the character image may be included outside this area. Therefore, the background image area binarization unit 13 binarizes the background image area and extracts a character image included in this area.

FIG. 7 is an explanatory diagram of the processing of the background image area binarization unit. In FIG. 7, first, the binarization threshold value of the background image area is calculated using the binarization threshold value of the character image area (S21). Next, a binarization process is performed on the background image area using the calculated binarization threshold (S22).

(Description of Binarization Example of Background Image Area) a: Method of Selecting a Representative Example from Character Image Area Binarization Threshold (First Method) As a first method, the character image area detection unit 11 In this method, a representative value is selected from the obtained threshold values (stored in a storage device or the like) of each area, and is applied to the entire background image area. For example, the lowest (black) gradation from the threshold value of each area is set as the threshold value of the background image area. If this threshold value is higher (white) than the fixed threshold gradation used in the character image area detection unit 11, the character image in the background image area that should have been blurred is binarized. It becomes more character-like.
As a method of selecting the threshold value, an average of the threshold values of the respective regions or the highest (white) gradation can be used.

FIG. 8 is an explanatory view of Example 1 of the binarization of the background image area.
It is. In FIG. 8, first, as shown in the above figure,
The character image area (the area within the dotted line) is
The area 1 and the area 2) are obtained. Next, as shown in the middle figure,
A character image area binarizing means in the character image area, and a single floor
Binarization is performed using a tonal image area binarization unit. At this time,
Let the binarization threshold in region 1 be t₁, Binarize in area 2
T_Two(T ₁> T_Two). Finally, in the figure below
As described above, of the binarization thresholds in the character image area,
Low tone_TwoPerforms binarization of the background image area with
Extract the part.

B: Method of adapting the threshold value of each area obtained by the character image area binarization unit to the background image area surrounding the area (second method) As a second method, the character image area 2 This is a method in which the threshold value of each area obtained by the value conversion unit is adapted to a background image area around the area. In this method, since the binarization threshold value obtained in a narrow region is used only around the narrow region, binarization for each character can be performed more appropriately.

FIG. 9 is an explanatory diagram of Example 2 of the binarization of the background image area.
It is. In FIG. 9, first, as shown in the above figure,
The character image area (the area within the dotted line) is
The area 1 and the area 2) are obtained. Next, as shown in the middle figure,
A character image area binarizing means in the character image area, and a single floor
Binarization is performed using a tonal image area binarization unit. At this time,
Let the binarization threshold in region 1 be t₁, Binarize in area 2
T_Two(T ₁> T_Two). Finally, in the figure below
As described above, an nxm area including each character image area (dashed line
Is binarized by the binarization threshold in each character image area.
You. That is, the n × m region including the region 1 is a binarization threshold value t.
₁And the n × m region including the region 2 is binarized by
Value t_TwoTo binarize. As a result, the image remains in the background image area.
The digitized character image portion is binarized. Note that the range of the area n × m
The box expands to a range that can be predicted to be characters.
is there. The result of binarization is shown in black.

As described in the above embodiment, the input multi-valued image is separated into a character image and an image region having other properties (background image region). , A binary threshold value of the background image area is determined from the binary threshold value at the time of the binary processing, and the binary image processing is performed on the background image area. Create For this reason, a binary image can be created with high accuracy and high speed even for an image in which the background of a character image, which has not been conventionally possible, has a colored background. The recognition accuracy can be improved by using the binary image.

(5): Description of installation of program Image binarizing means 10a, image recognizing means 14a, multi-value image binarizing section 10, character image area detecting section 11, character image area binarizing section 12, background The image area binarizing unit 13, the binary image recognizing unit 14, and the like can be configured by a program, and can be configured by a main control unit (CP
U) is executed and stored in the main memory. This program is generally processed by a computer. This computer is configured by hardware such as a main control unit, a main memory, a file device, a display device, and an input device such as a keyboard.

The program of the present invention is installed on this computer. This installation is performed on portable recording (storage) media such as floppy disks, magneto-optical disks, etc.
These programs are stored and installed in a file device provided in the computer via a drive device for accessing a recording medium provided in the computer or via a network such as a LAN. You. Then, the program steps necessary for the processing are read out from the file device to the main memory, and are executed by the main control unit.

[0061]

As described above, the present invention has the following effects.

(1): The input multi-valued image is separated into a character image region and a background image region by an image binarizing means, and each of the separated character image regions is subjected to a binarization process. The binarization threshold value of the background image area is determined from the binarization threshold value at the time of the binarization processing, and the binarization processing of the background image area is performed. Even if there is, binarization can be performed with high accuracy, and binarization can be performed at high speed because the binarization process is performed with a fixed threshold value.

(2): The image binarization means selects a low gradation representative value included in the input multi-valued image and performs binarization processing.
In order to separate the created binary image into a character image region containing a large number of character images and a background image region other than the character image region, only the black portion having a low gradation is binarized to easily separate the character image region. be able to.

(3): When the image data is binarized by the image binarizing means, when the character image area of the separated character image has a single gradation, the background image is included in the character image area of the single gradation. Since the character image area is changed and the binarization process is performed, even if the entire character image area is a character image, it is possible to prevent the character image from being output thinly and perform appropriate binarization. it can.

(4): The image binarization means finds each binarization threshold value in the binarization processing performed on each of the separated character image areas, and includes the binarized area. Since the area within the rectangle thus enlarged is subjected to the binarization processing using the obtained binarization threshold value, the character image area that has not been extracted can be appropriately binarized as the character image area.

(5): The input multi-valued image is separated into a character image area and a background image area, and each of the separated character image areas is subjected to binarization processing. Image binarization means for determining a binarization threshold value of the background image region from a binarization threshold value and performing binarization processing of the background image region;
An image recognizing means for recognizing the binary image created by the binarization process and a computer-readable recording medium on which a program for causing a computer to function are recorded. By installing the document image recognition device, it is possible to easily provide a document image recognition device that can binarize a character image with high accuracy even if the image has a colored background, and that can binarize it at high speed. .

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is an explanatory diagram of a configuration of a document image recognition device according to an embodiment.

FIG. 3 is an explanatory diagram of processing of a character image area detection unit according to the embodiment.

FIG. 4 is an explanatory diagram based on an image of a binary image area separating unit in the embodiment.

FIG. 5 is an explanatory diagram of a character image area binarizing unit according to the embodiment.

FIG. 6 is an explanatory diagram of an example of a binarization process using a discriminant criterion in discriminant analysis according to the embodiment;

FIG. 7 is an explanatory diagram of processing of a background image area binarizing unit according to the embodiment;

FIG. 8 is an example 1 of binarization of a background image area in the embodiment.
FIG.

FIG. 9 illustrates an example 2 of binarization of a background image area according to the embodiment.
FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input multivalued image 10a Image binarization means 14a Image recognition means

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B029 AA01 CC29 DD07 EE17 5C077 LL18 MP05 PP27 PP28 PQ08 RR02 RR16

Claims (5)

[Claims] 1. A document image recognizing apparatus comprising: an image binarizing unit for generating a binary image from an input multi-valued image; and an image recognizing unit for recognizing the generated binary image. The binarizing means separates the input multi-valued image into a character image region and a background image region, performs a binarization process on each of the separated character image regions, and performs a binarization threshold in the binarization process. A document image recognition apparatus, wherein a threshold value for binarizing the background image area is determined from a value, and a binarization process is performed on the background image area. 2. The image binarizing means selects a low gradation representative value included in the input multi-valued image and performs a binarizing process.
2. The document image recognition apparatus according to claim 1, wherein the generated binary image is separated into a character image region including many character images and a background image region other than the character image region. 3. The image binarizing means according to claim 1, wherein when the separated character image area is entirely a single tone of the character image, the character image area of the single tone includes a background image. 2. The document image recognition apparatus according to claim 1, wherein the character image area is changed to perform a binarization process. 4. The image binarizing means obtains each binarization threshold value at the time of the binarization process performed on each of the separated character image regions, and includes the binarized region. 2. The document image recognition apparatus according to claim 1, wherein the area within the rectangle thus enlarged is binarized by the obtained binarization threshold value. 5. An input multi-valued image is separated into a character image region and a background image region, a binarization process is performed on each of the separated character image regions, and a binarization threshold in the binarization process. Image binarization means for determining a binarization threshold value of the background image area from the values and performing binarization processing of the background image area; and an image for recognizing the binary image created by the binarization processing. A computer-readable recording medium that stores a program for causing a computer to function as a recognition unit.