JPH11328306A - Method and device for extracting logical element of document image, and record medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately extract bibliographic items by using not a character recognition result, but only its layout information at the time of recognizing logical structure. SOLUTION: A layout feature extracting means 103 divides an input document image into elements such as areas and detects features regarding document layout structure. A logical model detecting means 104 detects a model matching the type of a document to be processed out of plural models and an element extracting process means 105 extracts bibliographic items from the document image by using the detected logical model. When it is decided (106) that the model need not be updated, the extracted bibliographic items are outputted (107) and when it is updated, the logical model is updated by using the decided document, a sample document, etc.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a method and an apparatus for extracting a logical element of a document image for extracting a logical element of the document image, and a recording medium storing a logical element extraction processing program.

[0002]

2. Description of the Related Art For example, when constructing an electronic library, a document filing system, a database, etc., it is required to recognize the logical structure of a document image and automatically extract bibliographic items, There is also a need for a technique for recognizing a folder and automatically classifying the folder into an appropriate folder.

Conventionally, as this type of document processing apparatus, there is a technique described in, for example, Japanese Patent Application Laid-Open No. 8-287189. In this device, a user sets a bibliographic item extraction rule in advance, and recognizes a document structure by matching with a plurality of character string patterns set in advance. However, in the above-described apparatus, the cost for character recognition is high, and it is necessary to recreate the extraction rules for each document.

[0004]

As another method for recognizing the logical structure of a document, there is an apparatus and a method for recognizing a logical structure of a document and a content of a document described in Japanese Patent Laid-Open No. 5-159101. In the above-described apparatus and method, the relationship between the elements of the document image and the consistency with the structural model are checked, and the contents of the sheet of the document image are recognized as the attribute parameters of the logical structural elements of the corresponding structural model. For this purpose, document structure recognition is performed using a graph structure model that links the document elements to nodes and the arrangement relation between the elements, but such a structure model is defined by the relative positional relationship between all elements. Therefore, there is a problem that it takes a lot of time to create the same, and when one certain element is erroneously recognized, there is a high possibility that erroneous recognition will occur for all other remaining element recognitions.

SUMMARY OF THE INVENTION It is an object of the present invention to accurately extract a bibliographic item by using only layout information without using a character recognition result when recognizing a logical structure, and to automatically identify a document type and to automatically extract a bibliographic item. A method for automatically selecting a matching logical model, further adding a plurality of sample document images and updating the logical model, thereby extracting bibliographic items without depending on layout variations,
A device and a recording medium are provided.

[0006]

According to the present invention, a document is input as a digital image, layout characteristics of the document are detected from the document image, and a plurality of logical models are obtained. , A logical model that matches a layout feature of the input document is detected, and a logical element is extracted from the document image using the detected logical model.

[0007] The invention according to claim 2 is characterized in that when the layout feature of the extracted logical element fluctuates by a predetermined threshold value or more, the logical model is updated using the document image. .

According to the third aspect of the invention, means for inputting a document as a digital image, means for detecting layout characteristics of the document from the document image, and layout characteristics of the input document from a plurality of logical models. Means for detecting a logical model that matches the logical model, means for extracting a logical element from the document image using the detected logical model, and the detected logical model based on a layout feature of the extracted logical element. And a means for updating the logical model using the document image when it is determined that the logical model is to be updated.

According to the present invention, a function of inputting a document as a digital image, a function of detecting a layout characteristic of the document from the document image, and a layout characteristic of the input document are selected from a plurality of logical models. A function of detecting a logical model that matches the logical model, a function of extracting a logical element from the document image using the detected logical model, and a function of the layout of the extracted logical element fluctuating by a predetermined threshold value or more. The computer-readable storage medium stores a program for causing a computer to implement the function of updating the logical model using the document image.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be specifically described below with reference to the drawings. In the present invention, bibliographic items are automatically extracted from a document image. For this reason, setting by the user as a logical model merely gives a bibliographic item name and a simple attribute to an element to be extracted, and a logical model can be easily created as compared with the conventional method.

In addition, since each element is extracted independently, it is not necessary to define the entire structure in the logical model. Therefore, the influence of erroneous extraction on a certain element on extraction of another element is very small.

For a document of the same document type, the number of lines, the character size, the absolute position, etc. of the bibliographic items vary greatly for each document. By preparing a plurality of document images for learning, It is possible to automatically determine the reliability of the layout feature and update the parameter of the reliability to create a logical model. Since the learning is performed only on the elements having fluctuations, the adverse effect on the extraction accuracy of other elements is very small. Therefore, a stable extraction result that does not depend on a change in a specific layout feature can be obtained.

[0013] The quality of the updated logical model largely depends on the document selected. For example, when a logical model is updated by learning a document having an exceptionally large variation in a document type, the logical model cannot be said to model the document type, and conversely, there is almost no layout variation. Adding a document does not increase the learning effect. In the conventional method, it is necessary to manually select a document suitable for updating the logical model. In the present invention, a document suitable for updating the logical model is automatically selected. In addition, since the result of character recognition is not used, processing is simplified, and elements can be extracted without depending on the character direction or language.

Embodiment 1 FIG. 1 shows the structure of an embodiment of the present invention. In the figure, 101 is an input document, 102 is an image input means for inputting a document as a digital image,
Reference numeral 103 denotes a layout feature extraction unit that divides an input document image into a plurality of elements, and extracts features related to a document layout structure from the divided elements and the like. 104 denotes a logic that detects an appropriate logical model for a document to be processed. Model detection means; 105, an element extraction processing means for extracting a logical element from a document image using the logical model; 106, a logical model update determining means for determining whether to update the logical model;
107 is an extraction result output means for outputting the extraction result, 108
Is a document database holding the extraction results, 109 is a sample image for creating a logical model, 110 is a logical model creating means for creating a logical model by designating bibliographic item names and attributes to elements to be extracted from the sample image, and 111 is a logical model A logical model updating unit 112 for updating the logical model in order to improve the accuracy of the model is a logical model management database 112 for managing the logical model.

FIG. 2 is a processing flowchart of the present invention. Hereinafter, the operation of the present invention will be described with reference to FIG.

The image input means 102 reads the input document 101 and obtains a digital document image (step 201).
Next, the layout feature extraction unit 103 divides the input document image into regions and lines, and cuts out characters in order to determine the character size and character spacing (step 202). Further, features relating to the document layout structure such as the coordinates of elements such as regions and lines, the size of characters, indents, fonts, and column information are obtained (step 203). The input document image may include a plurality of pages. For example, when the input document is a dissertation, the input document is composed of a cover and a text, and the bibliographic information is usually described on the first page. Therefore, there is no need to process other body pages. Also, the input document image may have margins at the top, bottom, left and right depending on the input condition. However, since the relative position information with respect to the print area constituting the document is used as the coordinate values of the divided elements, Adverse effects can be avoided. Further, the line direction of the document is determined in order to accurately correspond to a vertically written document, but it is not necessary to recognize characters. The division result of the document image can be saved in memory,
Alternatively, the data may be written out to a file (for example, a code file with a tag).

Subsequently, the logical model detecting means 104 detects a logical model that matches the document type to be processed from a plurality of different logical models registered in the logical model management database 112 (steps 204 and 205). .
If there is no matching logical model, the input document is determined to be an undefined document (step 211). If the number of registered logical models is one, or if the document type to be input is known in advance, the process of step 204 may not be performed. Also, when the user can specify the logical model directly to the system,
Step 204 can be omitted. Further, even when a plurality of documents having different formats are successively input, the logical model detecting means 104 automatically selects the optimum logical model, so that the user can view all the documents without being aware of the input document format. It can be processed collectively.

FIG. 3 is a flowchart of a process for detecting a logical model. The logical model is detected by detecting layout features specified in the logical model (steps 302 and 30).
This is performed using the similarity between 3) and the layout feature of the processing target document (step 304). When the similarity is equal to or more than a certain value (step 305) and the difference between the similarities of the first candidate and the second candidate is sufficiently large (step 306), it is determined that the logical model is to be obtained (step 307). .
In addition, according to the similarity of the first candidate, a first criterion serving as a criterion is determined.
The threshold value of the similarity difference between the candidate and the second candidate may be dynamically changed.

Returning to FIG. 2, the element extraction processing means 105 extracts an area corresponding to bibliographic items from the document image by using the logical model detected by the logical model detecting means 104 (step 206). FIG. 4 is a diagram illustrating extraction of bibliographic items by matching with a logical model. The bibliographic item area stored in the model dictionary is matched with each candidate area of the document image. That is, for the bibliographic matter area and each candidate area, the similarity is calculated for each layout feature (area position, character size, number of lines, etc.), and the candidate area having the higher similarity (for example, from the first candidate to the third candidate). ) To extract.

Subsequently, the logical model update judging means 106
Determines whether to update the logical model described above (step 207). As a result of the determination, if the information is not updated (step 208), the extraction result output means 107 outputs the candidate area of the bibliographic item extracted as described above to the document database 108 (step 209). Update of the logical model (step 210) will be described later.

Next, the logical model creating means 110 will be described. FIG. 5 is a processing flowchart for creating a logical model. The logical model creation means 110 performs the processing on the model document image divided into a plurality of elements (step 501,
502), the user gives a bibliographic item name to the element to be extracted (step 503), and then gives an attribute (step 504). As the attribute, for example, the maximum number of rows of the element is given. Attributes are not necessarily specified because they are treated as auxiliary features. The logical model is created by combining the bibliographic item name, the attribute, and the layout feature of the designated area, and outputs the logical model to the logical model management database 112 (step 505). This logical model does not require a document structure, for example, a structure for graphing between elements, and is therefore easy to create and maintain.

FIG. 6 shows an example of a logical model. The logical model is described by a code with a tag as shown in FIG. 6 and stored in the logical model management database 112. In FIG. 6, in the first line, in the case of SGML (grammar describing the logical structure of a document), DTD (document type definition) and the like are defined. The second line describes information about the entire document, such as the print area range and the number of areas. Document data is actually described from the third line. The <area> tag describes, as attributes, user-designated information such as a bibliographic item name, a maximum number of lines, the number of lines, a coordinate value, a character size, a font value, and a character pitch, and a layout feature.

Next, the logical model updating means 111 will be described. If a document type with large intra-document variations is to be processed, this logical model should sufficiently cope with intra-document variations even if a logical model is extracted using a logical model created from a single sample image. Can not. In such a case, if the logical model is updated by newly adding a sample document and re-executing using the logical model, a logical element extraction result with higher accuracy than before the update is obtained. A method for automatically selecting a sample image suitable for updating the logical model has not yet been proposed. In the present invention, the logical model update determination unit 111 can automatically select a sample image suitable for updating the logical model.

FIG. 7 is a flowchart of the logical model update determination process. The logical model update determination unit 106 extracts a logical element from the processing target document using the logical model (Step 703), and detects the similarity (Step 704). At this time, even if the similarity of the input document is equal to or greater than a predetermined value, if there is a candidate area having a large variation in the layout characteristics of the input document, by updating the logical model using this input document, It is determined that a more accurate logical model can be created and updated (step 70)
5, 706). Otherwise, it is determined that the effect of updating the logical model is small (step 707).

The logical model updating unit 111 updates the logical model using the document determined by the logical model updating determining unit 106 or the sample document directly specified by the user. FIG. 8 is a processing flowchart of the logical model update. A logical model to be updated and a sample document image for addition are prepared (steps 801 and 802). The additional sample document image is divided using the same processing procedure as that of the logical model creating means 110 (step 803), layout characteristics are obtained, and a bibliographic item name is given (step 804).
Quantitative and qualitative variations in layout characteristics are detected for each bibliographic item (steps 805 and 806). If the variation is equal to or greater than a threshold value set for each layout feature, the corresponding part in the logical model is updated. And outputs it to the logical model management database 112 (step 807).

FIG. 9 is a diagram for explaining a logical model update. For example, in updating the character size feature, the allowable range of the character size has a width. In the update of the font feature, if the feature value defined in the logical model is different from the font of the additional sample image, the font feature is redefined to a value 0 designating that the font is not used for the logic extraction process. There are two types of feature parameters updated in this way: quantitative parameters and qualitative parameters. Coordinates, character size, and character spacing features are quantitative parameters, and fonts and appearance frequency features are qualitative. Parameter.

<Embodiment 2> As in the above embodiment, when bibliographic items are extracted using models prepared for each document type, the type of the input document is automatically identified, and an appropriate model is automatically identified. The choice is very effective. On this occasion,
If the bibliographic item extraction model and the document identification model are shared, resources can be effectively used and the model can be easily managed.

As an example of using a ruled line when identifying a document, there is a document system described in Japanese Patent Application Laid-Open No. 7-144622. In this system, vertical and horizontal ruled lines are extracted from a document image, and sheet identification (document identification) is performed using position information and length information of the ruled lines and identification information for each style (document type) in an identification dictionary. .

However, the above-described system is difficult to apply to a document having no ruled line. In addition, when the ruled line is blurred and cut off in the middle, when the input document contains many charts, or when a document image is input, a large amount of noise is generated. In the case where it exists, the possibility of erroneous recognition in the ruled line matching process is high. In other words, the document identification ability of the above system largely depends on the accuracy of extracting ruled line information.

In this embodiment, not only a fixed form document in which absolute coordinates and dimensions of a bibliographic item such as a form are defined (that is, a format is defined), but only the bibliographic item and its order are defined. Such a document (that is, a format is determined) is also processed, and the type of the document is identified with high accuracy by using bibliographic items and ruled line matching, which are invariable information of the document.

That is, in the present embodiment, one or more logical element models are created for each document type in order to automatically identify the type of the document that is randomly input, and the document is invariant information. A document type is identified with high accuracy by using bibliographic items and ruled line matching. Therefore, what the user sets as a logical element model is to give an identifier and a simple attribute to a bibliographic area defined as a style, and the logical element model can be easily created as compared with the conventional method. it can.

In the bibliographic item matching, since each element is processed independently, it is not necessary to define the structure of the entire document in the logical element model. Therefore, the influence of a processing error on a certain element on the whole is reduced. further,
Since the character recognition result is not used, the processing is simple, and the document type can be identified without depending on the line direction (vertical and horizontal writing) and the language.

The second embodiment includes two processes, a document identification process and a logical element model creation process. FIG. 10 shows the second embodiment.
Is shown. In FIG. 10, reference numeral 1001 denotes an input document;
1002, an image input unit for inputting a document as a digital image; 1003, an image dividing unit for dividing an input document image into a plurality of elements and extracting layout information;
A document identification unit 1004 selects an optimal model by comparing an input document with a plurality of logical element models.
005 is a result output means for outputting the result of the document identification means,
1006 is a document database holding a document identification result;
1007, a plurality of model documents for constructing a logical element model; 1008, an image input unit for inputting a model document image; 1009, an input model document image is divided into a plurality of elements to extract layout information The image dividing means 1010 designates a bibliographic item name and an attribute for a necessary element for a plurality of types of model documents divided into a plurality of elements, and a logical element model creating means 1011 for creating a logical element model. This is a logical element model management database that holds a logical element model.

FIG. 11 is a flowchart of the document identification process. Hereinafter, the document identification processing will be described. The image input unit 1002 reads the input document 1001 and obtains a digital document image (step 1101). The image dividing means 1003 divides the input document image into elements, and cuts out a character area, a ruled area, a character line, and a character (step 110).
2) Then, layout features related to the document layout structure such as element coordinates, character size, indent, font, and column information are obtained (step 1103). In addition, the line direction of the document is obtained in order to accurately correspond to a vertically written document, but it is not necessary to recognize characters.

The document identification means 1004 collates the input document with a logical element model prepared in advance for each document type, and calculates the inter-document distance between the input document and the model (step 1104). The method for calculating the inter-document distance will be described with reference to FIG.

In accordance with the inter-document distance, a threshold for document identification cutoff is calculated (step 1105). The method of calculating this threshold will be described later.

If there is only one model having a distance value equal to or smaller than the threshold value, the model is selected as the correct answer, the result is output, and the processing is terminated (steps 1106 and 110).
7, 1110). If there are a plurality of models having distance values equal to or smaller than the threshold value, these are determined to be rejected, and the process is terminated there (steps 1106 and 1110).
08, 1109, 1110) or whether to select only one optimal model in the detailed identification processing (step 1)
111), any of which is performed in accordance with the user's specification.

FIG. 12 is a processing flowchart for creating a logical element model. Hereinafter, the creation of the logical element model will be described with reference to FIG. Image input means 1008
First, a sample document 1007 serving as a model is read, and a model document image is input (step 1201). Next, the image dividing means 1009 divides the document image into elements,
A character area, a ruled line area, a character line, and a character are cut out (step 1202), and layout features related to the document layout structure such as element coordinates, character size, indent, font, and column information are detected (step 1203).

In the logical model creating means 1010, the user designates a bibliographic item name for the bibliographic item element of the divided document (step 1204) and an attribute (step 1205). The logical element model is stored in the logical element model management database 10
11 (step 1206). FIG. 6 shows an example of a logical element model.

Next, a method of calculating the inter-document distance between the input document and the logical element model will be described with reference to FIG. The inter-document distance is a bibliographic item distance value e using matching of a bibliographic item region specified by the user (steps 1303 and 13).
04) and the ruled line matching distance value 1 (steps 1305, 1306) using the vertical and horizontal ruled line information (e + 1) (step 1037).

A method of calculating a bibliographic item distance value and a ruled line matching distance value between an input document and a logical element model will be described with reference to FIG.

(Method of calculating bibliographic item distance value) The number of bibliographic items specified by the user in the logical element model is N, and the number of feature dimensions used for matching is M (step 1402,
1404). First, bibliographic items ai (i = 1,..., N) specified in the model from among the elements of the input document
Are extracted (step 140).
3). This is x (ai) (i = 1,..., N).
The element extraction is performed by using, for example, the method described in the first embodiment.

The element x (ai) is compared with the layout feature of the bibliographic item element ai corresponding to each, and a feature matching distance value from the first dimension feature to the M-th feature is calculated (step 1405). . For example, in the case of a character size feature, the feature matching distance value is a difference in character size between comparison elements.

The feature matching distance value of the j-th feature of the i-th bibliographic item ai is represented by fi (j). When feature matching distance values are obtained for all features, an M-dimensional feature matching distance value vector (fi (1),..., Fi
(M)) is obtained (step 1407).

At this time, the matching distance value e (i) between the document and the i-th bibliographic item ai is obtained by the following equation as the sum of each layout feature matching distance value and a predetermined weight αj (step 1408).

E (i) = α1fi (1) + α2fi
(2) + ... + αMfi (M) Here, αj is a constant determined corresponding to the j-th layout feature, and is used for normalizing each layout feature matching distance value.

When the matching distance value between the document and the document is calculated for all the bibliographic items, a distance value d by bibliographic item matching between the document and the model is calculated by the following equation (step 1410). That is, it is obtained as a sum of ei and a predetermined weight βi.

E = β1ei +... + ΒNeN Here, βi is a weight determined by the relative coordinates with respect to the printing range of the bibliographic items specified in the model. It has been empirically known that bibliographic items appearing at the end of a document have an important meaning as bibliographic item matching. Therefore,
The weighting coefficient βi takes a larger value as the bibliographic item appears near the end of the document, and takes a smaller value as the bibliographic item appears near the center of the document.

The ruled line matching distance value is calculated in the same manner as the above-described bibliographic item distance value calculation. However, features that do not exist in the ruled line area, such as character size and character pitch, are excluded from the processing. Further, the distance value based on the relative coordinates is set more strictly than in the case of calculating the bibliographic item distance value.

An example of a threshold for document identification cutoff will be described. Assuming that the inter-document distance between the input document and the first candidate model is d, a function thr (d) for determining a threshold is defined by the following equation. In this example, thr (d) depends only on the inter-document distance between the input document and the first candidate model.

Thr (d) = C1 + C2d (C1 and C2 are constants) If the inter-document distance between the input document and the first candidate model is larger than a threshold value, it is determined that there is no optimal model for the input document.

The detailed identification processing in step 1111 of FIG. 11 will be described. The basic processing flow is almost the same as the document classification based on the large classification described above. Hereinafter, only different parts will be described. In the detailed identification, the distance between the input document and the logical element model is the bibliographic item distance value using the matching of the bibliographic item region specified by the user and the ruled line matching distance value using the vertical and horizontal ruled line information, as in the large classification. Is defined as the sum of

On the other hand, in a document, the order of appearance of bibliographic items is guaranteed to be always constant. Therefore, in the detailed identification, when calculating the bibliographic item distance value, the bibliographic item regions are sequentially matched from the upper left of the model document (in the case of vertical writing, from the upper right), and their appearance ranks are taken into consideration.
Elements that have been included in the bibliographic item candidates in the past are excluded, or an order restriction is imposed, taking into account the continuity of the area, and supplementing appropriately if there is a gap in the middle. Using the inter-document distance obtained in this way, more detailed identification is finally performed.

As described above, according to the second embodiment, the document identification processing is hierarchized, and the similarity between the layout features corresponding to the bibliographic items and ruled lines of the document and the layout features of the model document is checked. Document type can be identified.
Further, since various layout features are detected, bibliographic items can be stably extracted with high accuracy from various input documents. Further, the number of dimensions of the layout feature can be generalized, so that it is possible to flexibly cope with a new addition or update of the layout feature. In addition, a logical element model having a high degree of freedom can be created, and therefore, a document type can be identified with high accuracy and stability. Furthermore, since a logical element model is created using a plurality of document samples for each document type, it is possible to cope with fluctuations in specific layout characteristics, to accurately and stably identify a document type, and to realize various layout characteristics. Document identification in consideration of the balance of Further, the document can be identified with high accuracy without depending on the variation of a specific layout feature. Further, since the distance (similarity) between the input document and the logical element model is obtained by using matching of bibliographic items and ruled lines, which are invariant information of the document, the document can be identified with high accuracy. When finding the distance (similarity) between the input document and the logical element model, bibliographic items and ruled line matching are performed independently of each other, so that the document can be identified by simple processing. Further, it is possible to cope with an error in the image dividing means. In addition, when obtaining the distance (similarity), by giving an appropriate weight to the matching result of bibliographic items and ruled lines, it is possible to identify a document with high accuracy. Furthermore, when matching bibliographic items or ruled lines, a proper weight is assigned according to the nature of each layout feature, so that document identification can be performed with high accuracy.

<Embodiment 3> The present invention is not limited to the above-described embodiment, but can be realized by software. When the present invention is realized by software,
As shown in FIG. 15, a computer system including a CPU, a memory, a display device, a hard disk, a keyboard, a CD-ROM drive, and a scanner is prepared.
-A computer-readable recording medium such as a ROM stores a program for realizing the function of extracting a logical element of a document image according to the present invention. Document images and the like are stored on a hard disk or the like. When the program is started, the document image data is read, a logical element extraction process is performed, and bibliographic items, which are logical elements extracted from the document image, are output to a display or the like.

[0056]

As described above, according to the present invention, a logical model appropriate for a document to be processed is automatically selected from a plurality of logical models, and a document image is selected by using the selected logical model. Bibliographic items can be stably extracted from Further, since the logical model is updated by the logical model updating means having a learning function, documents having various layout characteristics can be handled, and bibliographic items can be extracted with high accuracy.

[Brief description of the drawings]

FIG. 1 shows a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a processing flowchart according to the first embodiment of the present invention.

FIG. 3 is a processing flowchart of logical model detection according to the first embodiment.

FIG. 4 is a diagram illustrating extraction of bibliographic items by matching with a logical model according to the first embodiment;

FIG. 5 is a processing flowchart of creating a logical model according to the first embodiment.

FIG. 6 shows an example of a logical model.

FIG. 7 is a flowchart illustrating a logical model update determination process according to the first embodiment.

FIG. 8 is a processing flowchart for updating a logical model according to the first embodiment.

FIG. 9 is a diagram illustrating updating of a logical model according to the first embodiment.

FIG. 10 shows a configuration of a second exemplary embodiment of the present invention.

FIG. 11 is a flowchart illustrating a document identification process according to the second embodiment.

FIG. 12 is a processing flowchart of creating a logical element model according to the second embodiment.

FIG. 13 is a flowchart illustrating a process of detecting an inter-document distance according to the second embodiment.

FIG. 14 is a processing flowchart of bibliographic item matching according to the second embodiment.

FIG. 15 shows a configuration of Embodiment 3 of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 101 Input document 102 Image input means 103 Layout feature extraction means 104 Logical model detection means 105 Element extraction processing means 106 Logical model update determination means 107 Extraction result output means 108 Document database 109 Sample image 110 Logical model creation means 111 Logical model update means 112 Logical model management database

Claims (4)

[Claims] 1. A document is input as a digital image, layout characteristics of the document are detected from the document image, and a logical model that matches the layout characteristics of the input document is detected from a plurality of logical models. A method for extracting a logical element of a document image, comprising extracting a logical element from the document image using the detected logical model. 2. The document according to claim 1, wherein when the layout feature of the extracted logical element fluctuates by a predetermined threshold value or more, the logical model is updated using the document image. How to extract logical elements of an image. 3. A means for inputting a document as a digital image, a means for detecting layout characteristics of the document from the document image, and a logical model that matches a layout characteristic of the input document from a plurality of logical models. Means for detecting, means for extracting a logical element from the document image using the detected logical model, and whether or not to update the detected logical model based on layout features of the extracted logical element And a means for updating the logical model using the document image when it is determined that the document image is to be updated. 4. A function for inputting a document as a digital image, a function for detecting layout characteristics of the document from the document image, and a logical model that matches the layout characteristics of the input document from a plurality of logical models. A function for detecting, a function for extracting a logical element from the document image using the detected logical model, and a function for extracting the logical element when the layout feature of the extracted logical element fluctuates by a predetermined threshold value or more. A computer-readable storage medium storing a program for causing a computer to realize a function of updating the logical model using an image.