KR101571681B1 - Method for analysing structure of document using homogeneous region - Google Patents

Abstract

The present invention relates to a method for analyzing a document structure by using a homogenous region which includes the steps of: binarizing a document image into black and white by analyzing the document image; extracting a homogenous region by dividing the document image with respect to at least a part of white lines based on the width of the white lines and the width of black lines which are formed by binarizing the document image; classifying connected components into text components and non-text components by analyzing the connected components included in the homogeneous region, and then obtaining a text document including the text components and a non-text document including the non-text components; dividing the text document based on the distance between the text components; and classifying the non-text components included in the non-text document based on feature values of the non-text components. Accuracy for classifying the connected components into the text components and the non-text components can be increased by analyzing the connected components based on the homogenous region.

Description

[0001] METHOD FOR ANALYSIS STRUCTURE OF DOCUMENT USING HOMOGENEOUS REGION [0002]

The present invention relates to a method of analyzing a document structure using a homogeneous region, and more particularly, to a method of extracting a homogeneous region through repeated division and analyzing a document structure based on the extracted homogeneous region.

Optical character recognition (OCR) refers to conversion of an image in the form of text into computer-readable text data through a predetermined character recognition process. Optical character recognition is widely used when a paper document or an electronic image including text is converted into an electronic document because the user can omit the manual operation of typing and inputting the text. For optical character recognition, a process of analyzing the structure of a document to be converted is required. By analyzing the structure of the document, the document can be classified into a text area, a table area, an image area, and a line area. The accuracy of the process of analyzing the document structure can affect the accuracy of optical character recognition.

Conventionally, a top-down method and a bottom-up method are mainly used for analyzing a document structure. The top-down method is a method of dividing a document image into blocks, dividing a block into text lines, and classifying text lines into words. The top-down approach can shorten the computation time and efficiently analyze the formal document image, but it has a disadvantage in that it is difficult to accurately divide the document image if the structure of the document image becomes complicated. On the other hand, the bottom-up method is a method of determining words included in a document image, merging words into text lines, and merging text lines into blocks. The bottom-up method is capable of analyzing a document image even if the structure of the document image is complicated. However, it has a disadvantage that it is difficult to appropriately set a threshold value necessary for analysis and the computation time is excessively long. Although a combination of a top-down scheme and a bottom-up scheme has been proposed to overcome the disadvantages described above, this scheme does not fundamentally solve the above-mentioned disadvantages.

Therefore, it is required to develop a method and apparatus for accurately analyzing the structure of a document image with a short computing time and a complexity.
[Related Technical Literature]
1. A method, system and computer-readable recording medium for recognizing a character string included in a document using a language model and an OCR (Patent Application No. 10-2008-0103890)

SUMMARY OF THE INVENTION It is an object of the present invention to provide an image processing apparatus and a method of processing a document image by dividing a document image repeatedly using characteristics of a white line and a black line included in a document image to thereby extract a homogenious region having the same characteristics And a method of analyzing the document structure.

Another problem to be solved by the present invention is to analyze a connected component based on the extracted homogeneous region, thereby to identify a homogeneous region that can correctly classify connected components into text components and non-text components And to provide a method for analyzing the structure of a document.

Another object of the present invention is to provide a method of analyzing a document structure using a homogeneous region capable of quickly and accurately classifying connected components into a text component and a non-text component by classifying connected components through two filtering operations will be.

The problems of the present invention are not limited to the above-mentioned problems, and other problems not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a method of analyzing a document structure using a homogeneous region, the method comprising: analyzing a document image to binarize the document image in monochrome; Extracting the homogeneous region by dividing the document image based on at least a portion of the white lines based on the width and the width of the black line; analyzing the linked components contained in the homogeneous region to determine the associated component as a text component and a non- Dividing a text document based on a distance between text components, and dividing the text document based on the property values of the non-text component, Non-text components included in the non-text document And a step of classifying the image data.

According to another aspect of the present invention, the step of extracting the homogeneous region includes the steps of obtaining a curve in the horizontal axis or the longitudinal axis direction of the document image indicating the frequency of the black pixels in the pixels of the document image, Determining a presence or absence of a heterogeneous region based on a statistical value of a frequency, determining whether or not a heterogeneous region exists, determining, based on a width of the white line and a width of the black line included in the heterogeneous region, And dividing the document image based on the document image.

According to still another aspect of the present invention, there is provided an image processing method including the steps of acquiring a curve, determining whether a heterogeneous region is present, and dividing a document image until a heterogeneous region does not exist do.

According to another aspect of the invention, acquiring a text document and a non-text document comprises analyzing the linked components to extract non-textual components of the connected components, separating the non-textual components from the homogeneous region, And storing the non-text document including the text component.

According to still another aspect of the present invention, extracting the non-text component comprises extracting a non-text component from a text component or a text component based on a property value including a coordinate, a width, a height and an area of a box surrounding the connected component and the connected component Non-text component.

In accordance with another aspect of the present invention, there is provided a method for extracting a non-text component, the method comprising: extracting a non-text component; and storing the non- And storing the text document including the text document.

According to still another aspect of the present invention, classifying the non-text component is characterized by classifying the non-text component into a line component, a table component, a separator component, and a noise component.

According to another aspect of the present invention, the step of dividing the text document is a step of dividing the text document based on the distance between the text components after inflating the text component through a morphology operation .

According to another aspect of the present invention, a document image, a text document and a non-text document are characterized by the same size.

According to an aspect of the present invention, there is provided a document distributor in a document structure using a homogeneous region, the document analyzing apparatus comprising: a document analyzing unit for analyzing a document image to binarize a document image in black and white, Extracting the homogeneous region by dividing the document image based on at least a part of the white lines based on the width of the black line, analyzing connected components included in the homogeneous region to classify the connected components into text components and non-text components, Obtaining a text document including a text component and a non-text document containing a non-text component, dividing the text document based on a distance between the text components, and based on the property value of the non-text component, A set of instructions to classify non-text components contained in And it characterized in that.

According to an aspect of the present invention, there is provided a computer readable medium including a processor and a memory, the processor analyzing a document image to binarize the document image in monochrome, Extracting the homogeneous region by dividing the document image based on at least a portion of the white lines based on the width of the line and the width of the black line, analyzing the linked components contained in the homogeneous region, Obtaining a non-text document including a text document and a non-text component including a text component, dividing the text document based on a distance between the text components, and based on the property value of the non-text component To classify non-text components contained in non-text documents , The memory, the document image, the homogeneous area, the connected component, a text document and the non-storage is characterized in that at least one of a text document.

The details of other embodiments are included in the detailed description and drawings.

The present invention can accurately extract a homogeneous region having the same characteristics by repeatedly dividing a document image by using the characteristics of the white line and the black line included in the document image.

The present invention has the effect of improving the accuracy of classifying connected components into text components and non-text components by analyzing the connected components based on the extracted homogeneous regions.

The present invention can quickly and accurately classify connected components into text components and non-text components by repeatedly classifying the connected components by performing two repetitions of filtering.

The effects according to the present invention are not limited by the contents exemplified above, and more various effects are included in the specification.

1 is a schematic block diagram of an apparatus for analyzing a document structure using a homogeneous region according to an embodiment of the present invention.
2 is a flowchart illustrating a method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention.
3 is a flowchart illustrating a method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention.
4A to 4D illustrate an exemplary embodiment for processing a curve representing the frequency of black pixels in a method of analyzing a document structure using homogeneous regions according to some embodiments of the present invention.
5A and 5B illustrate homogeneous and heterogeneous regions in a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention.
6A and 6B illustrate an exemplary embodiment for partitioning a document image in a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention.
FIG. 7 is a conceptual diagram illustrating a relationship among connected components in a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention. FIG.
FIG. 8 exemplarily illustrates a text document and a non-text document obtained by a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention.
9 illustrates an exemplary document image analyzed by a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention.
10 is a graph illustrating the success rate of a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention, and the manner of achieving them, will be apparent from and elucidated with reference to the embodiments described hereinafter in conjunction with the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Is provided to fully convey the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims.

Although the first, second, etc. are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are used only to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may be the second component within the technical scope of the present invention.

Like reference numerals refer to like elements throughout the specification.

It is to be understood that each of the features of the various embodiments of the present invention may be combined or combined with each other partially or entirely and technically various interlocking and driving is possible as will be appreciated by those skilled in the art, It may be possible to cooperate with each other in association.

Various embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

1 is a schematic block diagram of an apparatus for analyzing a document structure using a homogeneous region according to an embodiment of the present invention.

Referring to FIG. 1, an apparatus 100 for analyzing a document structure using a homogeneous region includes a processor 110 and a memory 120. The apparatus 100 for analyzing the document structure using the homogeneous region may be a general purpose computer, a special purpose computer, a smart phone, a desktop or a notebook computer, or the like, or an accessory device used in combination with them.

The processor 110 performs various operations in an apparatus for analyzing a document structure using a homogeneous region. The processor 110 extracts the homogeneous region by binarizing the document image in black and white, divides the document image, and analyzes connected components included in the homogeneous region to generate a text document containing the text component and a non- Text components that are included in the non-text document based on the property values of the non-text component, and to classify the non- . The above-mentioned operation will be described later in detail with reference to FIG. 2 and FIG.

The memory 120 temporarily stores data and exchanges data with the processor 110. The memory 120 may temporarily store data for the document image, homogeneous area, connected components, text components, non-text components, text documents, and non-text documents described above.

Although not shown in FIG. 1, an apparatus for analyzing a document structure using a homogeneous region may include an input means such as a camera for acquiring a document image.

2 is a flowchart illustrating a method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention. For convenience of explanation, FIG. 1 will be described together.

A method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention is started by analyzing a document image and binarizing a document image in black and white by the processor 110 (S210). In order to perform a method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention, it is required to convert pixel values of a document image into pixel values representing white or pixel values representing black. The pixel value is data representing the brightness of the pixel. For example, a pixel value representing white may be 0, a pixel value representing black may be 1, a pixel value representing white may be 1, and a pixel value representing black may be 0. If the document image is a color image, the processor 110 may first convert the RGB components of the color image to a grayscale image of the color image. After converting the color image to a grayscale image, the processor 110 may convert the pixels having pixel values below the threshold value to 0 and convert the pixels with pixel values above the threshold value to 1 . The threshold value may be a predetermined value or a value determined differently depending on the pixel. For example, the threshold value may be determined using the Sauvola technique (Sauvola, J. and Pietikainen, M., 2000. Adaptive document image binarization. Pattern Recognition , 22 (2): 225-236. [Doi: 10.1016 / S0031-3203 99) 00055-2]).

Next, the processor 110 extracts a homogeneous region by dividing the document image based on at least a part of the white lines based on the width of the white line and the width of the black line of the binarized document image (S220) . The processor 110 may extract the homogeneous region by repeatedly dividing the binarized document image. Step S220 will be described later in detail with reference to FIG.

Next, the processor 110 analyzes a connected component included in the homogeneous region to classify the connected component into a text component and a non-text component, classifies the text component and the non- A non-text document containing the text component is obtained (S230). The processor 110 may obtain a text document and a non-text document by analyzing each of the extracted homogeneous regions separately to separate connected components. Step S230 will be described later in detail with reference to FIG.

Next, the processor 110 divides the text document based on the distance between the text components (S240). The processor 110 can extract a homogeneous text area having a uniform distance between the text components by dividing the text document based on the distance between the text comforts. The processor 110 may expand the text component through a morphology operation and then partition the text document based on the distance between the text components. For example, the processor 110 may expand each of the text components at a constant magnification, and then analyze the expanded text components to extract a homogeneous text area that includes text components that form a single entity. Text components that make up a unit are text components whose expanded text components make up a connected component. Processor 110 may expand the text components to different magnifications from each other. For example, the processor 110 may set the magnification for expanding the text component to be different depending on the characteristics of the homogeneous region including the text component. The processor 110 may set the magnification to be large when the distance between the text components in the homogeneous area including the text component is long and to set the magnification when the distance between the text components is close. By setting the magnification differently according to the characteristics of the homogeneous area, the homogeneous text area can be extracted more accurately.

Next, the processor 110 classifies the non-text components included in the non-text document based on the property values of the non-text component (S250). Processor 110 may classify non-text components into, for example, line components, table components, separator components, and noise components. The property value of a non-text component means the width, height, and area of the bounding box surrounding non-text and non-text components.

For example, the processor 110 may classify non-text components of very large or small aspect ratios and non-text components of very high density into line components. Processor 110 may classify non-text components that are low in density, include a rectangular component, a bounding box not passing through other text components, and a plurality of text components in the bounding box into table components. The processor 110 may classify non-text components that contain text components as separator components, such that the bounding box is very low in density (e.g., less than 0.01). The processor 110 may classify non-text components that are very small in area into noise components. The processor 110 may classify the line components, table components, separator components, and non-text components that remain after the noise components are sorted into image components. By performing the classification described above in a non-text document containing only non-text components, the accuracy of classification of non-text components can be improved.

3 is a flowchart illustrating a method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention. For convenience of explanation, FIG. 1 will be described together.

A method of analyzing a document structure using a homogeneous region according to an embodiment of the present invention is started by analyzing a document image and binarizing a document image in black and white by the processor 110 (S210). Since step S210 has been described with reference to FIG. 2, redundant description will be omitted.

Next, the processor 110 obtains a horizontal or vertical axis curve of the document image indicating the frequency of the black pixels among the pixels of the document image (S221). When the processor 110 first analyzes a document image, the processor 110 performs the following analysis on the entire area of the document image. 4A to 4D illustrate an exemplary embodiment for processing a curve representing the frequency of black pixels in a method of analyzing a document structure using homogeneous regions according to some embodiments of the present invention. The processor 110 may produce a histogram of the vertical projection of the black pixels of the document image from the document image shown in Fig. 4a, as shown in Fig. 4b. However, as shown in Fig. 4B, the calculated histogram is not suitable for analysis because it has a very irregular profile. Accordingly, the processor 110 can smoothly convert the profile of the histogram to obtain a curve in the longitudinal axis direction, as shown in Fig. 4C.

An exemplary equation for calculating a histogram of a document image projected on the horizontal axis and smoothening the calculated histogram is as follows.

[Equation 1]

&Quot; (2) "

Here, x is the horizontal axis coordinate of the pixel, y is the vertical axis coordinate of the pixel, a is the maximum value of the abscissa, and b is the maximum value of the ordinate. f (x, y) is a pixel value at coordinates (x, y) and has a value of 0 (white pixel) or 1 (black pixel). p _x is the number of black pixels with abscissa x and P is the set with p _x as an element. Here, s means a kernel size for smoothing, sp _x is a value obtained by converting p _x smoothly, and SP is a set having sp _x as an element.

As described above, the processor 110 projects a black pixel on a horizontal axis or a vertical axis, and smoothes the black pixel, thereby generating a horizontal axis or a vertical axis direction curve representing the frequency of black pixels according to the horizontal axis or vertical axis coordinate of the document image Can be obtained. Here, a curve means a set of consecutive points.

Next, the processor 110 determines whether a heterogeneous region exists based on the statistical values of the curve change rate and the black pixel frequency (S222). As shown in FIG. 4D, the processor 110 may calculate the rate of change of the curve by differentiating the obtained curve. Since the distribution of the black pixels can be grasped through the rate of change of the curve, the rate of change of the curve can be a criterion for judging the homogeneity of the region.

For example, the processor 110 may calculate a pole at which the rate of change of the curve is zero, calculate a distance between neighboring poles, and calculate a variance of the distance between poles. The low variance of pole-to-pole distance means that the analyzed region is homogeneous, and the high variance of pole-to-pole distance means that the analyzed region is a heterogeneous region. For example, the processor 110 may determine a region having a dispersion of pole-to-pole distances of 1.1 or less as a homogeneous region and a region having a pole-to-pole distance variance of 1.1 or more as a heterogeneous region. 4A to 4D, it can be seen that the illustrated text area is a homogeneous area because the distance between the poles is constant. 5A and 5B illustrate homogeneous and heterogeneous regions in a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention. Referring to FIG. 5A, the regions shown in FIG. 5A may be determined as homogeneous regions because the distances d1 and d2 between neighboring poles are constant. Referring to FIG. 5B, the region shown in FIG. 5B is different from the neighboring pole distances d3 and d4, so that the region shown in FIG. 5B can be determined as a heterogeneous region.

An exemplary equation for calculating the derivative of the curve, calculating the pole of the curve, calculating the distance between the poles, and calculating the variance of the distance between the poles is as follows.

&Quot; (3) "

&Quot; (4) "

&Quot; (5) "

&Quot; (6) "

Here, _x is the slope of the dsp sp _x on the abscissa x, DSP is the set of elements having a dsp _x. x _i is the abscissa of the pole at which the slope of sp _x is zero, and E is the set of elements with x _i as an element. de _i means the distance between neighboring poles, DE is a set with de _i as an element. V refers to the dispersion of the de _i, and μ indicates the average of the de _i and, n refers to the number of de _i.

Next, the processor 110 divides the document image based on at least a part of the white lines based on the width of the white line and the width of the black line included in the heterogeneous region when a heterogeneous region exists (S223). When the processor 110 analyzes the document image for the first time, the analysis is performed on the entire area of the document image, so that there is a high probability that the heterogeneous area exists. 6A and 6B illustrate an exemplary embodiment for partitioning a document image in a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention.

Referring to FIG. 6A, the processor 110 may calculate the widths (w1, w2, w3) of the white lines included in the heterogeneous region. The processor 110 may divide the heterogeneous region based on the white line having the largest width among the white lines included in the heterogeneous region and having a width greater than the median of the white lines included in the heterogeneous region. In FIG. 6A, the processor 110 divides the heterogeneous region with respect to the white line having a width greater than the intermediate value and having the largest width w2.

Referring to FIG. 6B, the processor 110 may calculate the widths (b1, b2, b3, b4) of black lines included in the heterogeneous area. The processor 110 determines a black line having a width greater than an intermediate value among the black lines included in the heterogeneous region and having the widest width among the black lines included in the heterogeneous region and sets the white line adjacent to the identified black line as a reference The heterogeneous region can be divided. In FIG. 6A, the processor 110 divides the heterogeneous region with respect to two white lines adjacent to the black line having a width larger than the intermediate value and having the largest width b3. The advantageous effect of the present invention is obtained that the homogeneous region can be extracted more accurately by dividing the heterogeneous region based on the width of the black line as well as the width of the white line.

The processor 110 may repeat the steps of acquiring a curve (S221), determining whether a heterogeneous region is present (S222), and dividing a document image (S223) until no heterogeneous region exists. That is, the processor 110 may divide the document image, and then perform an analysis based on each of the divided regions to determine whether each of the divided regions is a heterogeneous region, and divide the heterogeneous region again. If the processor 110 repeatedly divides and each of the divided areas is judged to be a homogeneous area, the processor 110 stops dividing the area.

The processor 110 may divide the document image in the horizontal axis direction by using the same method after the vertical axis direction division of the document image is completed. When the division in the horizontal axis direction is completed, the document image includes rectangular areas, all of which correspond to homogeneous areas. The following steps are performed based on the extracted homogeneous regions. The advantageous effect of the present invention is obtained that the analysis of the document structure can be performed more accurately by classifying the components included in the document after repeating the division of the document image as described above to extract the homogeneous region.

Next, the processor 110 obtains the text document and the non-text document by analyzing the connected components and extracts non-text components among the connected components (S231). A connected component is a set of black pixels connected together in a document image. For example, since the letter "ㅐ" is connected to all the black pixels, one connected component can be extracted from the letter "ㅐ", and the letter "ㅔ" &Quot; from which two connected components can be extracted. The processor 110 may extract the connected components from each of the homogeneous regions included in the document image and label the extracted connected components. The processor 110 may classify the connected components as text components or non-text components based on the property values, including the coordinate, width, height, and area of the bounding box surrounding the connected component and the connected component. A bounding box refers to a rectangle having a minimum area including sides connected to a horizontal axis and a vertical axis of a document image including a connected component. The bounding box can be defined as the coordinates of the upper left point and the lower right point. Processor 110 may extract non-textual components of connected components through two filtering operations.

The processor 110 may extract the connected components as non-text components that satisfy a predetermined condition as a first order filtering to roughly extract non-text components. Since the primary filtering is a process for quickly extracting connected components that are very likely to be non-text components, the setting of the conditions needs to be done carefully. For example, the processor 110 may include a connected component that is small enough to be visually perceptible, a connected component that has a very low density so that the interior is likely to be an empty shape, and a connected component that has an extremely high or low aspect ratio, You can extract it as a non-text component.

An exemplary equation for expressing the condition for extracting the non-text component in the primary filtering is as follows.

&Quot; (7) "

&Quot; (8) "

&Quot; (9) "

&Quot; (10) "

Here, CC _i is the number of the i-th and the connected component, the pixels are occupied by an area of a CC _i CC_area (CC _i) CC _i is contained in the homogeneous specific region, _area, and T denotes the threshold. For example, T _area may be six. B_CC _i denotes the bounding box of CC _i , Inside (B_CC _i ) denotes the number of bounding boxes of other connected components included in B_CC _i , and T _inside denotes a threshold value. For example, T _inside may be 3. B_dens (CC _i) means a value obtained by dividing the number of pixels occupied by the CC _i by the number of pixels occupied by the B_CC _{i i} and a density of the CC, and T means a _dens threshold. For example, T _dens can be 0.06. W _i is the width of B_CC _i , H _i is the height of B_CC _i , HW_ratio (CC _i ) is the smaller of W _i and H _i divided by the larger value, and T _HWratio is the threshold it means. For example, T _HWratio may be 0.06.

In the first order filtering, the conditions presented by equations (7), (8) and (10) are exemplary, and not limited thereto, various other conditions for extracting non-text components may be used.

Processor 110 may extract the remaining connected components that have not been extracted in the first-order filtering as non-text components through more precise second-order filtering than first-order filtering.

In the second filtering, first, the processor 110 may extract a connected component having a certain area or more as a candidate component. For example, the processor 110 may extract a connected component that is larger than a certain multiple of the median value of the area of all connected components included in the same homogeneous region, as a candidate component. Since the areas of the text components included in the same homogeneous region are not so different from each other, a connected component having a large area difference from other connected components within the same homogeneous region may correspond to a non-text component.

The processor 110 may calculate the distance between the bounding boxes of the candidate component and other connected components based on the coordinates of the bounding box of the extracted candidate component and the candidate component, You can extract components and calculate the distance between connected components that are adjacent in the horizontal direction. The text components contained within the same homogeneous region are adjacent to one connected component to the left and one connected component to the right, with a constant distance from each other. Based on the characteristics of the text components described above, the processor 110 may extract the non-text components based on the number of connected components that are adjacent in the horizontal direction to the candidate component and the distance between connected components that are adjacent in the horizontal direction to the candidate component .

An exemplary equation for extracting a candidate component in the second order filtering is as follows.

&Quot; (11) "

Here, CCs means a set of connected components included in a homogeneous region including a candidate component, CC _j means all connected components included in CCs, CC _i means a candidate component, and median (CC_area) means Means the middle value of the area of all connected components included in the CCs, and k means the threshold value.

k is a settable constant. If k is set large, the accuracy of extraction is improved. If k is set small, the computing time is shortened. An exemplary equation for calculating a suitable k value in consideration of both the accuracy and the computing time is as follows.

&Quot; (12) "

Here, mean (CC_area) means an average value of the area of all connected components included in the homogeneous region including the candidate component.

An exemplary equation for extracting a non-text component of a candidate component is as follows.

&Quot; (13) "

&Quot; (14) "

FIG. 7 is a conceptual diagram illustrating a relationship among connected components in a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention. FIG. Referring to FIG. 7, in Equations (13) and (14), LNN (CC _i ) means the connected component closest to CC _i and LNN (CC _j ) means connected component closest to CC _j , num {LNN (CC _i )} means the number of connected components closest to CC _i and leftmost, and max _2nd (CC_area (CC _i )) means the second largest area of CC _i . LNWS (CC _i) the CC _i and LNN the distance between (CC _i), and RNWS (CC _i) the CC _i and the distance between the RNN (CC _i), and RNN (CC _i) the CC _i and the right And MeanWS is the average of RNWS (CCi). Processor 110 may extract candidate components that satisfy one or more of Equations 13 or 14 above as non-text components.

By using the first filtering, which roughly extracts non-text components through a simple operation as described above, and the second filtering, which extracts all non-text components through precise operations, it is possible to accurately extract non-text components during a short computing time The advantageous effect of the present invention is obtained.

Next, the processor 110 separates the non-text component from the homogeneous region and stores the non-text document including the non-text component (S232). The processor 110 may separate the extracted non-text components and then collect the separated non-text components to obtain a non-text document that is a collection of non-text components, (120).

Next, the processor 110 determines whether there is a separated non-text component (S233). The processor 110 may determine whether a non-text component is present by determining whether the non-text component is extracted. For example, the processor 110 may calculate the pixel distribution of the document image before the non-text component is separated and the pixel distribution of the document image after the non-text component has been separated so that if the two are different, And judges that there is no separated non-text component if they are the same.

The processor 110 may repeat the step of extracting the non-text component (S231) and the step of separating the non-text component and storing the non-text document until the non-text component is not extracted. For example, the processor 110 may perform steps S231 and S231 until the pixel distribution of the document image before the non-text component is separated and the pixel distribution of the document image after the non- S232) can be repeated.

The processor 110 may collect the connected components, i.e., text components, remaining after the non-text component is removed, to obtain a text document that is a collection of text components, and store the obtained text document in the memory 120 have.

FIG. 8 exemplarily illustrates a text document and a non-text document obtained by a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention. Referring to FIG. 8, the processor 110 may obtain a binarized document image 720 by binarizing the document image 710, extract the homogeneous region by dividing the binarized document image 720, Extracts text components and non-text components from the binarized document image 720 by analyzing the image 720 by homogeneous regions, and generates a text document 730 containing only text components and non-text containing only non- Document 740 may be obtained. The document image 710, the binarized document image 720, the text document 730, and the non-text document 740 may be the same size.

Next, the processor 110 divides the text document based on the distance between the text components (S240), and the processor 110 determines whether the non-text component included in the non-text document (S250). Steps S240 and S250 may be performed simultaneously, sequentially, or in reverse order. Step 240 and step S250 have been described with reference to FIG. 2, and redundant description will be omitted.

9 illustrates an exemplary document image analyzed by a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention. 9, a text document is divided into homogeneous text areas 911, 912, 913, 414, 915, 916, 917, 918, 919 based on the distance between text components. The non-text component included in the non-text document is a table component 941 having a low density and containing text therein, a linear line component and separator components 931, 932, 933, 934, 936, , Line components, or image components 921, 922, 923, and 924 that are not included in the separator component. As shown in FIG. 9, each of the components of the document image can be accurately classified according to the type by the analysis method of the document structure using the homogeneous region according to an embodiment of the present invention.

10 is a graph illustrating the success rate of a method of analyzing a document structure using a homogeneous region according to some embodiments of the present invention. The success rate shown in Fig. 10 is a value calculated through " F-measure ". The success rates of Comparative Examples 1, 2, 3 and 4 shown in Fig. 10 indicate the success rates when using "Fraunhofer", "FineReader", "Tesseract" and "DICE", respectively.

Referring to FIG. 10, the classification success rate of the text component is 96.88% and the classification success rate of the non-text component is 92.27% when the document structure analysis method using the homogeneous region according to an embodiment of the present invention is used. Compared with the comparative examples 1, 2, 3 and 4, the classification success rate of the method of analyzing the document structure using the homogeneous region according to the embodiment of the present invention, 4, which is about 20% higher than that of Comparative Example 1. As described above, a good classification success rate for non-text components can be achieved by extracting homogeneous regions through repetitive segmentation and then classifying the non-text components by repeating the first-order filtering and the second-order filtering.

In this specification, each block or each step may represent a part of a module, segment or code that includes one or more executable instructions for executing the specified logical function (s). It should also be noted that in some alternative embodiments, the functions mentioned in the blocks or steps may occur out of order. For example, two blocks or steps shown in succession may in fact be performed substantially concurrently, or the blocks or steps may sometimes be performed in reverse order according to the corresponding function.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, which is capable of reading information from, and writing information to, the storage medium. Alternatively, the storage medium may be integral with the processor. The processor and the storage medium may reside within an application specific integrated circuit (ASIC). The ASIC may reside within the user terminal. Alternatively, the processor and the storage medium may reside as discrete components in a user terminal.

Although the embodiments of the present invention have been described in detail with reference to the accompanying drawings, it is to be understood that the present invention is not limited to those embodiments and various changes and modifications may be made without departing from the scope of the present invention. . Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. Therefore, it should be understood that the above-described embodiments are illustrative in all aspects and not restrictive. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

Method of analyzing document structure using 100 homogeneous regions
110 processor
120 Memory
d1, d2, d3, d4 Distance between adjacent poles
w1, w2, w3 Width of the white line
d1, d2, d3, d4 width of black line
710 document image
720 Binary document image
730 text documents
740 non-text documents
911, 912, 913, 414, 915, 916, 917, 918, 919,
921, 922, 923, 924 image components
931, 932, 933, 934, and 936 line components and a separator component
941 Table component

Claims (11)

Analyzing the document image to binarize the document image in black and white;
Extracting a homogeneous region by dividing the document image based on at least a part of the white lines based on a width of a white line formed by binarizing the document image and a width of a black line;
Classifying the connected component into a text component and a non-text component by analyzing a connected component included in the homogeneous region, and classifying the text component including the text component and the non- Obtaining an included non-text document;
Dividing the text document based on a distance between the text components; And
And classifying the non-text component included in the non-text document based on a property value of the non-text component. The method according to claim 1,
The step of extracting the homogeneous region comprises:
Obtaining a curve in the horizontal axis or the vertical axis direction of the document image indicating the frequency of black pixels among the pixels of the document image;
Determining a presence of a heterogeneous region based on a change rate of the curve and a statistical value of the frequency of the black pixel;
Dividing the document image based on at least a part of the white lines based on the width of the white line and the width of the black line included in the heterogeneous region when the heterogeneous region exists. A method for analyzing a document structure using homogeneous regions. 3. The method of claim 2,
Further comprising the step of obtaining the curve, determining whether the heterogeneous region is present, and dividing the document image until the heterogeneous region does not exist. An analysis method of document structure used. The method according to claim 1,
The step of obtaining the text document and the non-text document
Analyzing the connected component to extract the non-text component of the connected component;
And separating the non-text component from the homogeneous region to store the non-text document comprising the non-text component. 5. The method of claim 4,
Wherein extracting the non-text component comprises:
Classifying the connected component into the text component or the non-text component based on a property value including a coordinate, a width, a height and an area of the box surrounding the connected component and the connected component. A method for analyzing document structure using homogeneous regions. 5. The method of claim 4,
Extracting the non-text component and storing the non-text document until the non-text component is not extracted; And
And storing the text document in which the non-text component is removed and the text component is included. The method according to claim 1,
Wherein classifying the non-text component comprises:
And classifying the non-text component into a line component, a table component, a separator component, and a noise component. The method according to claim 1,
Wherein the step of dividing the text document comprises:
Expanding the text component through a morphology operation and then dividing the text document based on a distance between the text components. The method according to claim 1,
Wherein the document image, the text document and the non-text document are the same size. Analyzing the document image to binarize the document image in monochrome,
Extracting a homogeneous region by dividing the document image based on at least a part of the white lines based on a width of a white line and a width of a black line included in the document image,
Analyzing connected components included in the homogeneous region to classify the connected component into a text component and a non-text component, acquiring a text document including the text component and a non-text document including the non-text component ,
Dividing the text document based on a distance between the text components,
And a set of instructions for classifying the non-text component included in the non-text document based on a property value of the non-text component. A processor; And
Memory,
The processor
Analyzing the document image to binarize the document image in monochrome,
Extracting a homogeneous region by dividing the document image based on at least a part of the white lines based on a width of a white line and a width of a black line included in the document image,
Analyzing connected components included in the homogeneous region to classify the connected component into a text component and a non-text component, acquiring a text document including the text component and a non-text document including the non-text component ,
Dividing the text document based on a distance between the text components,
Classifying the non-text components included in the non-text document based on a property value of the non-text component,
The memory comprising:
Wherein the at least one of the document image, the homogeneous region, the connected component, the text document, and the non-text document is stored.

Images