JP2000341515A - Method and device for extracting text area and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To surely extract a text area that is printed in a colored ground from a colored image. SOLUTION: In a method for extracting a text area, a colored image expressed in R, G, and B colors is transformed into color coordinates, having high independency between components reflecting the visual characteristics of humans (S20) and image texture characteristics are extracted for every transformed component (S30). Then each pixel is classified into a text area and a non-text area by utilizing the image texture characteristics (S40) and the connecting components of the pixels classified into the text area are extracted as the text area (S50).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the field of document image processing and, more particularly, to a technique for extracting a text area composed of characters and symbols from a digital color image.

[0002]

2. Description of the Related Art Conventionally, documents are mainly printed in black and white, but recently, with the spread of color printers, the colorization of documents has been progressing. Many of the document image processing techniques target black-and-white binary images and gray-scale grayscale images, but the need for color document images is increasing.

As an application field of such a color document image processing technology, a digital color image obtained by inputting a document image printed on paper to a computer through an image input device such as an image scanner is processed, and the OCR is performed. Document image recognition that encodes text information such as characters, and image area separation for improving image quality by separating and processing character areas and medium tone areas from input images in digital copiers. No. In order to perform such processing, it is necessary to first extract a text area composed of characters, symbols, and the like from a digital color image.

[0004] As a conventional technique for extracting a text region from a digital color image, a document [Anil K. Jain and
Yao Chen, "Address block location using color and
texture analysis, "CVGIP: Image Understandin
g, vol. 60, no. 2, pp. 179-190, September 1994]. In this method, a luminance value of each pixel of a digital color image represented by three components of RGB is calculated, and a grayscale image using the luminance value (or using the G value as a luminance value) is generated. Each pixel value of this grayscale image is 2
By binarizing, it is separated into a white area and other areas. In addition, the image is separated into a text region and other regions using the texture characteristics of the grayscale image.
Then, the two processing results are integrated, and only the area surrounded by the white area is finally extracted as the text area from the text area by the image texture feature.

[0005]

The above prior art is effective for the purpose of extracting an area of a text printed on a white background. In addition, a digital F
Since a Gabor filter that can be implemented by hardware is used as the IR filter, the processing speed is excellent. However, the extraction accuracy of a text area printed on a color background is poor, and particularly, on a grayscale image, when the gray value of the ground (background) and the text are close, or the gray value of the ground is close to black. In such a case, there is a problem that normal text region extraction is difficult.

SUMMARY OF THE INVENTION It is therefore an object of the present invention to solve such a problem and to provide a method and an apparatus for accurately extracting text regions of various color document images.
Another object of the present invention is to enable text region extraction closer to human color perception.

[0007]

In order to achieve the above object, according to the present invention, an image texture feature is extracted for each component of a digital color image to be processed, and processing is performed using the extracted image texture feature. Classify each pixel in the target digital color image into a text area or a non-text area. Preferably, the digital color image to be processed is converted to another color coordinate having a high degree of independence between components reflecting human visual characteristics, and an image texture feature of each converted component is extracted. Is used to classify each pixel.

[0008]

Preferred embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, FIG. 1 is a flowchart showing an example of a text area extraction processing procedure, and FIG. 2 is a flowchart showing an example of an image texture feature extraction processing procedure. FIG. 3 is a block diagram illustrating an example of a text region extracting apparatus that executes the processing illustrated in FIGS. 1 and 2. Hereinafter, the contents of the text region extraction process and the configuration of the text region extraction device will be described in association with the flow of the process shown in FIG. 1 or FIG.

<< Step S10 >> Color image input unit 1
Reference numeral 00 captures a digital color image including a text area through a color image input device such as an image scanner or a digital copier, a communication line, or a recording medium such as a floppy disk or a magneto-optical disk. Here, the input digital color image will be described assuming that the color of each pixel is represented by color coordinates corresponding to R, G, and B. The input digital color image is stored in the memory 101.

<< Step S20 >> Color coordinate conversion unit 10
In step 2, the digital color image stored in the memory 101 is converted into another color coordinate that reflects human visual characteristics and has high independence between components, thereby obtaining R, G, and B colors.
The component is represented by another three components. When the values of the respective pixels are collected for each component of the converted color coordinates, three gray value images, that is, a first component image, a second component image, and a third component image are obtained. These component images are stored in memories 103 and 10.
4, stored in the memory 105, respectively. As the color coordinates after the conversion, for example,
Color coordinates consisting of two independent hue components are available.

(1) Pseudo KL space Conversion from RGB space to pseudo KL space is given by the following matrix.

(Equation 1) The first component I1 represents shading (luminance), and the second component I2 and the third component I3 represent hue.

(2) YyCxCz space A space obtained by linearly approximating the CIE Lab space, and the conversion from the XYZ space is given by the following matrix.

(Equation 2) The first component Yy represents shading (luminance), and the second component Cx and the third component Cz represent hue. Here, XY from RGB space
The transformation to Z space is given by the following matrix:

(Equation 3)

(3) Yy-rg-yb space Conversion from XYZ space is given by the following matrix.

(Equation 4)

Considering the above equation (3), YyCxCz and Yy-rg-yb are the opposite color responses of the second component of red-green,
The third component is a color space composed of components corresponding to the opposite color responses of yellow-blue respectively.

In addition to the above-mentioned conversion, conversion to a YIQ space (Y is luminance, I and Q are color differences) used in television can be used.

<< Step S30 >> In the image texture feature extraction unit 106, each of the memories 103, 104, 10
The texture feature of each component image (the gray-scale value image of each component) stored in 5 is extracted. Specifically, an image texture feature is calculated using a Gabor filter as shown in FIG.

That is, the value of the ith component (i = 1, 2, 3) at the pixel (m, n) is expressed as Ii (m, n), and
The image of the component (i = 1, 2, 3) is represented by Ii. The Gabor filter is a linear filter for detecting directional / periodic pattern features in an image, and is determined by four parameters (direction θ, spatial frequency u, smoothing parameters σx, σy). Given by the formula.

[0018]

(Equation 5) Here, the unit of the spatial frequency u is 1 / pixel. The smoothing parameters σx and σy are standard deviations (unit: pixel) of Gaussian functions in two orthogonal directions, respectively. I
The result of applying Gabor filter to i (m, n) is convolution Gi (m, n) = Ii (m, n) * h (m, n; u,
θ, σx, σy). FIG. 4 shows an example of the impulse response of the Gabor filter when θ = 0. Note that σx
And σy are set in accordance with the values of u and θ so that the frequency bandwidth and the directional bandwidth have appropriate values.

Referring to FIG. First, several values of u and θ are selected (step S31), and for each set of values of (u, θ), a filtered image Gi (m, n) = Ii, which is the result of applying a Gabor filter to each component image. (M,
n) * h (m, n; u, θ, σx, σy), i = 1, 2,
3 is calculated (step S32).

Next, the value t of each pixel of each filter image is converted by the following equation (step S33).

(Equation 6)

Each image composed of the converted pixel values is represented by r
(M, n), and as shown in the following equation, the pixel (m, n)
Calculates the image texture feature by window averaging the values of the image r (m, n).

(Equation 7) Here, Wm, n is a window of size M × M centered on pixel (m, n).

A vector of the image texture feature for each set of (u, θ) values calculated from the three component images in this manner is created and passed to the classification unit 107 (step S35).

<Step S40> The classifying unit 107 classifies each pixel into a text area or other non-text area using the image texture features extracted in the previous step. The classification result is stored in the memory 108.
For this classification, a method of pattern classification such as a nearest neighbor method or a neural network can be used. Using the learning data in advance, the classification unit 107 is trained based on the feature amounts of the pixels belonging to the text region and the feature amounts of the graphics belonging to the non-text region. In this case, the feature amount is a vector in which texture features calculated using a Gabor filter with various frequencies u and directions θ are arranged. When M values of the frequency u and N values of the angle θ are selected, the dimension of the feature vector is 3MN.

<< Step S50 >> The text area extraction unit 109 refers to the classification result of each pixel stored in the memory 108, and connects the connected components of the pixels classified into the text area to the text of the input digital color image. Extract as a region.

According to the above-described method and apparatus of the present invention, since the image texture characteristics of each component of the digital color image are used, a text area printed on a color background, which is a problem in the prior art, is used. Can be accurately extracted regardless of whether the grayscale value of the ground is close to the text when converted into a grayscale image or the grayscale value of the ground area is close to black. In particular, prior to the extraction of image texture features, since the conversion to color coordinates with high independence between components reflecting human visual characteristics is performed, it is a color that can be clearly distinguished from ground color according to human vision. However, in the RGB data, a text region printed in a color having a small difference can be reliably extracted, and the extracted text region can be brought closer to a text region recognized by a human. That is, it is possible to extract a text area closer to human color sense.

In the extraction of the image texture filter in step S30, it is not always necessary to use a Gabor filter. However, as described above, the Gabor filter can be easily implemented by hardware, and is therefore excellent in processing speed.

Although not shown, according to another embodiment, the color coordinate conversion in step S20 is omitted, and image texture features of R, G, and G components of the input digital color image are extracted, which are classified. It is used in step S40. Even in this case, it is possible to reliably extract a text area printed on a color background. However, as described above, it is possible to extract a text region closer to human color sense by performing the color coordinate conversion in step S20.

According to the present invention, for example, as shown in FIG.
U200, memory 201, display device 202, user input device 203 such as keyboard and mouse, hard disk device 204, medium drive device 206 for reading and writing recording medium 205 such as floppy disk, magneto-optical disk, CD-ROM, image It is also possible to use software such as a general computer in which external interface units 209 and 210 for connection to an image input device 207 such as a scanner and a communication line 208 are connected by a system bus 211.

A program for causing a computer to execute each step of the text region extraction processing of the present invention described with reference to FIGS. 1 and 2, or the text region extraction of the present invention described with reference to FIG. A program for realizing the function of each unit of the apparatus by using hardware of a computer is read from the recording medium 205 in which the program is recorded into the memory 201 and executed. Alternatively, the program is stored in the hard disk drive 20 in advance.
4 is read into the memory 201 and executed at the time of execution.

The digital color image to be processed is
For example, the recording medium 205 or the external interface unit 2
The image data is input to a computer via the image input device 207 or the communication line 208 connected via the input terminals 09 and 210. The text area extraction result data is obtained on the memory 201, which is stored in the hard disk drive 204 or the recording medium 205.
, The screen output to the display device 202, or the communication line 208.

[0031]

As is clear from the above description, according to the first to sixth aspects of the present invention, it is possible to extract a text area as if it were printed on a color background, and this is the most conventional technique. Even if the gray value of the ground and the text are close to each other when converted to a grayscale image and the gray value of the ground area is close to black, reliable text area extraction is possible. In addition, according to the second or fifth aspect of the present invention, it is possible to extract a text region with high accuracy close to human color sense. According to the invention of claim 3 or 6,
It is easy to speed up image texture feature extraction, which is advantageous for speeding up the entire text region extraction process.
Further, according to the seventh or eighth aspect of the invention, it is possible to obtain such effects that the above-described text region extraction processing can be easily performed using a general computer.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an example of a text area extraction processing procedure according to the present invention.

FIG. 2 is a flowchart illustrating an example of an image texture feature extraction processing procedure;

FIG. 3 is a block diagram illustrating an example of a configuration of a text region extraction device according to the present invention.

FIG. 4 is a diagram illustrating an example of an impulse response of a Gabor filter when a direction θ = 0.

FIG. 5 is a block diagram showing an example of a computer for implementing the present invention by software.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 color image input unit 101 digital color image storage memory 102 color coordinate conversion unit 103 first component image storage memory 104 second component image storage memory 105 third component image storage memory 106 image texture feature extraction unit 107 classification unit 108 Classification result storage memory 109 Text area extraction unit

Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (Reference) H04N 1/46 Z F Term (Reference) 5C077 MP06 MP08 PP01 PP27 PP31 PP32 PP43 PQ22 5C079 HB01 HB11 LA01 LA06 MA01 NA29 5C082 AA27 BA02 BA20 BA27 BA34 CA21 CA54 CB01 DA87 5L096 AA02 AA06 AA07 BA07 BA17 EA26 FA44 FA81 GA40 GA55 9A001 HH28 HH31 JJ28

Claims (8)

[Claims] 1. A method for extracting a text area of a digital color image to be processed represented by a certain color coordinate for each pixel color, the image texture feature for each component of the color coordinate from the digital image to be processed. And extracting each pixel of the digital color image to be processed into a text area or a non-text area using the extracted image texture feature. 2. A method for extracting a text area of a processing target digital color image represented by a certain color coordinate, wherein a color of each pixel is represented by human visual characteristics in the processing target digital color image. Conversion to other color coordinates with high independence between components is performed, and image texture features for each component of the other color coordinates are extracted from the converted digital color image, using the extracted image texture features. A method for extracting a text region, wherein each pixel of the digital color image to be processed is classified into a text region or a non-text region. 3. The method according to claim 1, wherein a Gabor filter is used for extracting image texture features. 4. A means for inputting a digital color image represented by a certain color coordinate of a color of each pixel, and means for extracting an image texture feature for each component of the color coordinate from the digital color image input by this means. Means for classifying each pixel of the input digital color image into a text area or a non-text area using the image texture features extracted by this means, and connected components of the pixels classified into the text area by this means Region extracting device, comprising: means for extracting a character as a text region. 5. A means for inputting a digital color image represented by a certain color coordinate of a color of each pixel, wherein the input digital color image has high independence between components reflecting human visual characteristics. Means for performing conversion to other color coordinates, means for extracting image texture features for each component of the other color coordinates from the digital color image converted by this means, and using the image texture features extracted by this means. Means for classifying each pixel of the input digital color image into a text area or a non-text area,
And a text region extracting apparatus comprising: a unit that extracts, as a text region, a connected component of pixels classified into a text region by the unit. 6. The text region extracting apparatus according to claim 4, wherein a Gabor filter is used for extracting image texture features. 7. A step of inputting a digital color image represented by a certain color coordinate for each pixel color, and extracting an image texture feature for each component of the color coordinate from the input digital color image in this step. Classifying each pixel of the input digital color image into a text area or a non-text area using the image texture features extracted by this step; and
A computer-readable recording medium on which a program for causing a computer to execute a step of extracting, as a text area, a connected component of pixels classified into a text area by this step is recorded. 8. A step of inputting a digital color image represented by a certain color coordinate for each pixel color, wherein in the input digital color image, independence between components reflecting human visual characteristics is obtained. Performing a conversion to other high color coordinates, extracting image texture features for each component of the other color coordinates from the digital color image after the conversion in this step, extracting the image texture features extracted in this step. Using a computer to perform a step of classifying each pixel of the input digital color image into a text area or a non-text area by using the computer and a step of extracting a connected component of the pixels classified into the text area from the step as a text area. A computer-readable recording medium on which a program for causing a computer to execute is recorded.