JP3850995B2 - Document image processing method and machine-readable recording medium storing a program for causing a computer to execute the document image processing method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
INDUSTRIAL APPLICABILITY The present invention is used in a document image recognition system that recognizes an input document image, and in particular, when extracting a character area from a document image, a document image processing method and document for correcting excessive integration of extracted character areas The present invention relates to a machine-readable recording medium recording a program for causing a computer to execute an image processing method.
[0002]
[Prior art]
Conventionally, in order to input a document image, perform character recognition processing from the document image, and acquire various layout information, it is necessary to perform region extraction processing such as a character region, a graphic region, and other regions.
[0003]
As a reference technical document for performing such region extraction processing, “Print Document Recognition System AutoReco / 2 -Image Process” (Information Processing Society of Japan 47th (late 1993) National Conference 2-97-98) is disclosed. Has been. Here, in particular, column boundary lines are detected, and character string elements are integrated into a region (this is referred to as the first prior art).
[0004]
Further, a technique for detecting the column state of the entire page and extracting the character region in accordance with the state is disclosed in “Document Image Region Division Method and Column Type Discrimination Method” of Japanese Patent Laid-Open No. 9-44594 ( This is called the second prior art).
[0005]
[Problems to be solved by the invention]
However, the conventional techniques as described above have the following problems. First, in the first prior art, over-integration may occur when the caption of a figure and the text are close to each other in a newspaper or the like. A specific example is shown in FIG. In FIG. 7, since the caption 704 is close to the text, it is erroneously extracted as one region in the form of 703.
[0006]
Also, in the method disclosed in the second prior art, when the character elements are dense such as in newspapers, a clear column is not detected at the initial stage, and the character area is obtained by integrating neighboring character elements. Will be extracted. That is, excessive integration occurs when close to each other as shown in FIG.
[0007]
Even in the method disclosed in the first prior art, the adjacent character elements are grouped in the initial stage, so that the edge portion of the step is not detected. Even in such a case, if the edge portion of the step is detected, a number of pseudo edges are detected in addition to the original edge, and the extraction of the character area becomes complicated as a whole.
[0008]
The present invention has been made in view of the above, and an object of the present invention is to correctly extract a character area even when characters belonging to different character areas are close to each other when extracting a character area from a document image. To do.
[0009]
[Means for Solving the Problems]
To achieve the above object, in the document image processing method according to claim 1, character elements in a document image are integrated to generate a line, and the generated line is integrated to extract a character area. In the document image processing method for correcting line integration for extracting the character area, a column information extraction step for extracting column position information from the character area, and an appearance frequency of the area width of the character area. a stage width estimation step of estimating a step width based, the step of extracting the correction candidate region based the previous SL-stage set of position information on the estimated variable width of a character region corresponding to the stage width and the estimated, the And an area correction step of dividing the correction candidate area using a forced division position corresponding to the position of the step in the corresponding character area.
[0010]
The document image processing method according to claim 2 is the document image processing method according to claim 1, wherein the region correction step is a step estimated from the position information of the step extracted in the column information extraction step. When a candidate area that should belong to the same stage as the area that matches the width is detected, and characters or character string elements in the area are merged into the character area again while referring to the position of the stage, the above-mentioned stage is followed. Assuming a forced division position, a character area consisting of a direction perpendicular to the direction of the candidate area that should belong to the same stage as the area that matches the extracted step width is generated outside this division position, and then The character area is evaluated .
[0012]
According to a third aspect of the present invention, there is provided a machine-readable recording medium in which a program for causing a computer to execute the document image processing method according to the first or second aspect is recorded.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, a document image processing method of the present invention and a machine-readable recording medium that records a program for causing a computer to execute the document image processing method will be described in detail with reference to the accompanying drawings.
[0014]
(System configuration)
FIG. 1 is a block diagram showing the configuration of a system to which a document image processing method according to an embodiment of the present invention is applied. In the figure, reference numeral 101 denotes an image input unit for optically reading and inputting a document including a document as a binary image using a CCD-equipped scanner or a facsimile machine. Further, as a specific means of the image input unit 101, a document image may be acquired from another device via a network (by a data communication unit 110 described later).
[0015]
When capturing an image via a network, for example, a wide-area data resource (access information on a WWW server) distributed on the Internet using a browser function (WWW (world wide web) browser function) prepared in advance is used. Browse data in languages such as HTML (Hyper Text Markup Language: document description language for the Internet information service using hypertext) specified by the URL (Uniform Resource Locator) .
[0016]
Reference numeral 102 denotes an image reduction unit that reduces an image input by the image input unit 101, 103 denotes a basic element extraction unit that extracts a character element from the image reduced by the image reduction unit 102, and 104 denotes a character that extracts a character area. An area extracting unit 105 is a column detecting unit that detects a column from the character area extracted by the character area extracting unit 104, 106 is a character area correcting unit that executes a character area correcting process described later, and 107 is the entire system. Is a control unit that performs overall control based on a predetermined control program, 108 is a data storage unit for storing various data such as input document data and layout information, 109 is a data communication path, 110 is, for example, , TCP (Transmission Control Protocol) / IP (Internet Protocol) protocol A configuration data communication unit to perform network communication according to.
[0017]
(System operation)
Next, the document image processing operation in the system configured as described above will be described. FIG. 2 is a flowchart showing the procedure of the document image processing method according to the embodiment of the present invention. First, a target document image is input by the image input unit 101 (S201). Further, the input document image is reduced by the image reduction unit 102 (S202).
[0018]
In this case, for example, by performing OR compression of 8 × 8 pixels into one pixel, the black pixels constituting the character are connected in the character string direction. The circumscribed rectangle of this connected component is extracted and used as a basic element. This basic element may be a single character or a part of a character string, and a table or the like alone becomes one large basic element.
[0019]
Subsequently, the basic elements are classified into characters, tables, diagrams, and the like, and the character elements are integrated to generate a line, and the lines are integrated to extract a region (S203). The flow of this process has been previously proposed by the present inventor in Japanese Patent Laid-Open No. 9-44594.
[0020]
Subsequently, column information is extracted (S204). This column structure will be described with reference to FIGS. The length of the extracted character area in the row direction is defined as “area width”, and the length in the direction perpendicular to the “area width” is defined as “area height”. Here, a histogram of “region width” is created for those having “region height” equal to or greater than a preset threshold value.
[0021]
That is, in the case of FIG. 3, 301 to 304 are character areas having a height equal to or higher than the threshold. This histogram is shown in FIG. In the histogram of FIG. 4, if there is a class having a high ratio of frequency to the total frequency, it is set as “step width”. In FIG. 4, a class having a width of 90 to 120 corresponds to this. In FIG. 3, 301 to 303 correspond to this.
[0022]
At this time, a restriction may be placed on the range that the “step width” can take. For example, in the example shown in FIG. 4, if 80 to 170 are within the range, even if the frequency of 30 to 60 classes is high, the step width is not adopted, and the step width is undecided finally. In the method disclosed in Japanese Patent Laid-Open No. 9-44594, a matched column state is detected from the distribution of character elements, and the detected column information may be used.
[0023]
If the column information can be extracted, the character area is corrected using the column information (S205). The details of the character area correction processing are shown in the flowchart of FIG.
[0024]
In FIG. 5, first, an area matching the estimated step width is detected (S501). In the example shown in FIG. Subsequently, a candidate area that should belong to the same stage as the area detected in step S501, that is, a correction target candidate is detected (S502). In this case, an area where the upper end or lower end of the area is aligned and the width does not change so much is a candidate area. In addition, when a fusion of captions is assumed, one of the conditions is that a diagram area exists in the vicinity. In the example shown in FIG.
[0025]
If a candidate area is detected as described above, next step S503 is executed. That is, the character area is extracted again using the character elements constituting the candidate area. At this time, a forcible division position (see FIG. 6) along the stage is assumed, and line generation and area generation are executed in a direction perpendicular to the reference line direction outside the position. FIG. 6 is an enlarged view of the corresponding portion of FIG. Rows 602 and 603 are generated using 601 elements outside the stage with respect to the forced division position. Then, the degree of matching in the generated row is calculated (S503).
[0026]
Subsequently, when the degree of matching is equal to or higher than a preset threshold value, this area is made formal, and the other character area is generated using the character elements inside the remaining forcible division positions (S504). ). As the degree of line alignment, it is possible to use the slope of the regression line of the center of gravity of each character element or the variance of the distance between elements. In addition, when a plurality of rows are generated outside the stage as shown in FIG. 6, a value that is inversely proportional to the difference in the height of the row (vertical direction in FIG. 6) may be used as the degree of matching.
[0027]
Therefore, by the processing operation as described above, an accurate region can be extracted even when the caption shown in FIG. 7 is close to the main caption.
[0028]
Incidentally, the document image processing operation in each of the embodiments described above is executed by the system shown in FIG. 1, but in addition to this, the document image processing method is recorded on a machine-readable recording medium as software and is executed on a computer. You may make it perform.
[0029]
【The invention's effect】
As described above, according to the document image processing method of the present invention (Claims 1 and 2 ), the position of the stage extracted in the column extraction process is referred to, and the excessively integrated character area is evaluated / For correction, even a document image in which characters belonging to different character areas are close to each other can be extracted as a character area that should be extracted by avoiding excessive integration of the character areas.
[0030]
Further, in the machine-readable recording medium according to the present invention (claim 3), by recording a program for executing a document image processing method according to the computer to claim 1 or 2, according to claim 1 or The operation described in 2 can be realized by a computer.
[Brief description of the drawings]
FIG. 1 is a block diagram showing the configuration of a system to which a document image processing method according to an embodiment of the present invention is applied.
FIG. 2 is a flowchart showing a procedure of a document image processing method according to the embodiment of the present invention.
FIG. 3 is an explanatory diagram showing an example of column information extraction according to the embodiment of the present invention.
FIG. 4 is a histogram example of a region width created and used when extracting column information according to an embodiment of the present invention.
FIG. 5 is a flowchart showing a detailed processing procedure of region correction processing in FIG. 2;
6 is an explanatory diagram showing an enlargement of FIG. 7 for an example of forced re-region extraction / evaluation according to the embodiment of the present invention.
FIG. 7 is an explanatory diagram showing a case where the caption of the figure and the text are close to each other.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 101 Image input part 102 Image reduction part 103 Basic element extraction part 104 Character area extraction part 105 Column detection part 106 Character area correction part 107 Control part 108 Data storage part 110 Data communication part

Claims (3)

In a document image processing method for integrating character elements in a document image, generating a line, extracting a character area by integrating the generated line, and correcting line integration for extracting the character area,
A column information extraction step for extracting column column position information from the character region;
A step width estimation step of estimating a step width based on the appearance frequency of the region width of the character region;
A step of extracting the correction candidate region based the previous SL-stage set of position information on the estimated variable width of a character region corresponding to the stage width and the estimated,
An area correction step of dividing the correction candidate area using a forced division position corresponding to a position of a step in the corresponding character area;
A document image processing method comprising: The region correction step detects a candidate region that should belong to the same step as the region that matches the step width estimated from the step position information extracted in the column information extraction step, and a character or character string in the region When the elements are merged into the character area again with reference to the position of the step, the forced division position along the step is assumed, and outside the division position, the same step as the region that matches the extracted step width is assumed. The document image processing method according to claim 1 , wherein a character area composed of a direction perpendicular to a direction of a candidate area to belong to is first generated, and then the character area is evaluated . A machine-readable recording medium having recorded thereon a program for causing a computer to execute the document image processing method according to claim 1 or 2 .

Images