JP2004240488A - Document managing device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document integrally managing system for paper/electronic file which is improved so that an operator can retrieve a document not from a date when the document has been scanned but from a date when the document has been actually prepared or a date close thereto on which the contents of the text are reflected. <P>SOLUTION: This document managing device for storing and managing an original image or an electronic document read by a scanner etc. is provided with a word outputting means for performing the character recognition of a character string in the original image by using a dictionary for recognition, and for performing morphemic analysis by using the word dictionary, and for outputting words being the recognition candidates, an extracting device for extracting a portion expressing dates or time from the word string outputted by the word outputting means, an estimating means for estimating a period when the original image has been prepared from the data related with the dates or time extracted by the extracting means and an imparting means for imparting the period estimated by the estimating means as the attributes of the original image. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a document management server that stores and manages a large number of documents on a computer, and in a situation where a document using a conventional paper medium and a digitized document are mixed in an office or various other situations, both can be used. The present invention relates to a system that is seamlessly integrated, and is devised so that usability is improved for needs such as classification, processing, and search.
[0002]
[Prior art]
In the document management server, in order to store and manage old paper documents on a new computer, scan the original with a scanner and divide the area for each type of image, such as characters, charts, photographs, etc. The text is converted into a character code string by OCR or the like, and the other parts are separately managed as image files. Generally, when a document is managed as a file on a computer, some additional information is stored as attributes of the document together with the content of the file itself. For example, the size and format of the file, the date and time when the file was created (or updated), the creator, and the like.
[0003]
By the way, once a paper document has been digitized, it is desired to perform a seamless search without any particular awareness of the distinction between documents in a document group that has already been created by input from a keyboard or the like and stored in a document storage device. Is likely to occur to the operator at some stage. Searches are most likely to be based on the content or format of the document, but tend to be based on the date the document was created. This is because the contents and the writing time of some documents are closely related.
[0004]
However, in the conventional system, when a paper document is digitized in the manner described above, the date and time at that time, that is, the date and time when the work of digitization is completed, is added as an attribute of the file. Was normal. Since paper documents have become the mainstream in offices and usually exist before the operation of a document management server for managing a large number of digitized documents in the office, it has existed for a long time. However, even if the date at the time of the reading operation is given an attribute as the date and time of creation of the file, it seems that the background is that the user wants to search using the original date of creation when searching.
[0005]
Looking at past technologies related to document date management and automatic generation of date data, according to the invention described in Japanese Patent Application Laid-Open No. 09-062665, “Document Management Apparatus”, the date information related to an atypical document contains the contents of the document. For the purpose of solving the problem that the distribution of date information related to the document over a certain period could not be easily grasped only by actually reading the document, It has been proposed that the frequency of a document related to a date is arranged from a plurality of time perspectives such as year / month / day of the week / day and presented in an easy-to-understand manner. The present invention overlaps with the present invention in that date data is extracted from the text, and is an excellent invention from the viewpoint of a GUI (Graphical User Interface). However, a document existing in paper is regarded as an electronic file. For coexistence, there is no viewpoint that the document management server operates essential data for managing a document on a computer called a file attribute.
[0006]
Also, the invention in Japanese Patent Application Laid-Open No. 06-103268 “Document management device” relates to a technology for detecting omission of description of a document and automatically embedding information that can be automatically generated in the document. Since it means the date when the document is registered, it can be uniquely determined from the value of the timer of the system, so it is obtained and inserted into the document.
[0007]
However, it has not been proposed to infer the actual creation date of a paper document created in the past from the text. For this reason, it is impossible to cope with old paper documents and digitized documents, and to manage dates suitable for searching.
[0008]
[Problems to be solved by the invention]
As mentioned above, if you want to search based on the date of creation from a mixture of paper documents scanned and digitized after the operation of the document management system and documents created as digitized data from the beginning However, in paper documents, the date and time of scanning and digitization are given as file attributes instead of the date actually created, so the date management standards differ from those of documents that have already been digitized. However, it can be said that it is difficult to find both of them uniformly, and there is a problem that the date and time of the scan has no relevance to the written contents and cannot be relied on at the time of the search.
[0009]
Accordingly, the present invention provides a method for searching a character recognition result and finding a character string portion that seems to contain the date when the document was actually created when scanning and digitizing a paper document. Is added as a file attribute, so that the operator can search by the date created or the date close to the date reflecting the contents of the text, instead of the date and time of scanning, when searching. The purpose is to provide an integrated document management system.
[0010]
[Means for Solving the Problems]
In order to achieve the above object, the present invention provides a document management device that stores and manages document images and digitized documents read by a scanner or the like, performs character recognition using a dictionary for recognizing character strings in the document images. A word output unit that performs a morphological analysis using a word dictionary and outputs a word as a recognition candidate thereof, and an extraction device that extracts a portion representing a date and time from a word string output by the word output unit, Estimating means for estimating the time when the document image was created from data on the date and time extracted by the extracting means, and providing means for giving the time estimated by the estimating means as an attribute of the document image. It is characterized by comprising.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
[0012]
FIG. 1 is a block diagram showing the overall configuration of a document management apparatus according to the present invention.
[0013]
In FIG. 1, reference numeral 101 denotes an OCR (optical character reading device) which reads a text portion in a document image and outputs a character code of a recognition candidate character. Reference numeral 102 denotes a morphological analyzer that divides the recognized character candidates output from the OCR 101 into morphemes while referring to the word dictionary 103. Reference numeral 103 denotes a word dictionary which is referred to when the morphological analysis device of the above 102 cuts out a word string from a candidate character string. Reference numeral 104 denotes a date data search device that searches the morphological sequence divided by the morphological analyzer 102 for data corresponding to date data.
[0014]
Reference numeral 105 denotes an attribute assigning device for writing some supplementary information as attributes of the original in addition to the contents when the read original image is stored as an electronic file. The attributes include, for example, the size and format of the created file, the date and time when the file was created (or updated), and the creator (in this case, an image reading operator). Reference numeral 106 denotes a document storage device, which includes a large-scale storage device such as a hard disk for storing digitized documents.
[0015]
Hereinafter, the processing flow will be described in three stages (1) to (3).
(1) Character recognition processing
Now, it is assumed that the document image shown in FIG. 3 has been input, and the flow of processing will be described according to the flowchart of FIG.
[0016]
First, the OCR device 101 performs character recognition on a text portion of a document image (step 201). As a result, a character code string including a plurality of character candidates is output for each read character. FIG. 4 shows character candidates output as a result of performing character recognition on 11 characters of... Date and time: June 20 (...) In the document image of FIG.
[0017]
Characters up to the fourth character candidate are shown for each recognized character, but in general the number of candidate characters to be output is not determined. The morphological analyzer 102 considers all combinations of these candidate characters, and divides them into possible morpheme strings with reference to the word dictionary 103 (step 202). FIG. 5 is an example of the division obtained by performing the morphological analysis of the portion shown in FIG.
(2) Genre estimation processing
Next, the genre estimating process in step 212 will be described.
[0018]
In order to estimate the time when the document was created, it is assumed that the approximate genre of the document has been determined. For example, in the case of a document such as a report or a plan document created in an office, the format is often fixed, and it is easy to find the creation date. On the other hand, in the case of general sources such as newspaper articles, travelogues, narratives, etc., the creation date is often not explicitly stated, and the date part is searched from the text to estimate the time of creation There is a need to. Here, a description will be given of a process of estimating the genre or category from the contents described in the main text when the genre or category cannot be determined based on the document format.
[0019]
For that purpose, it is important that a large number of digitized documents accumulated in the past have already been classified into genres and categories. Therefore, first, a method using a vector space model will be described as one of methods for classifying a document group stored in the document management server into one of predetermined categories according to a user's intention. In this vector space model, words, documents, and categories useful for classification are represented by vectors, and the category to which the documents belong is determined using the direction of the vectors as a guide. The processing of automatic document classification using this vector space model can be divided into a learning phase and a classification phase.
[0020]
In the learning phase, words useful for classification (hereinafter, referred to as effective words) are selected from learning documents that have been correctly classified in advance, and each effective word is expressed as a vector. This vector is called an effective word vector, and the components of this effective word vector are obtained from the appearance frequency, the word co-occurrence probability, and the like. Each effective word is used as a heading, and the effective word vector is stored as an effective word dictionary. Also, by calculating the weighted average of each effective word vector included in the learning document, the contents of the document can be expressed in a vector, and by averaging the document vectors belonging to each category, the characteristics of the category can be expressed. The calculation of the folder vector is performed.
[0021]
In the classification phase, a document to be classified is expressed as a vector using the effective word dictionary obtained in the learning phase (hereinafter, referred to as a document vector), the document vector is compared with a folder vector, and according to the comparison result, The category to which the document to be classified belongs is determined.
[0022]
As shown in FIG. 6, the genre estimating apparatus includes a learning document database 601 that holds learning documents, a classification target document holding unit 602 that stores classification target documents, and valid words that extract valid words from the learning documents. An extraction means 603, an effective word dictionary 604 for holding the extracted effective words together with the degree of belonging to each category, an evaluation item table 618 in which importance values are described for each evaluation item for evaluating the importance of the effective words, An effective word number calculation unit 605 for obtaining the number of effective words included in each document by referring to the learning document and the effective word dictionary, and an effective word number holding unit 606 for holding the obtained effective word number in each document. And a co-occurrence frequency calculating means 607 for calculating a co-occurrence frequency of each set of effective words based on the number of effective words, and an effective word vector calculating means 609 for obtaining an effective word vector of each effective word by referring to the co-occurrence frequency And learning Document vector calculating means 611 for obtaining a document vector by referring to an effective word vector for each of a document and a classification target document, and folder vector calculating means for obtaining a folder vector of each category using the document vector obtained for a learning document 613, a category estimating unit 615 that compares the document vector obtained for the classification target document with the folder vector of each category, and estimates a category to which the classification target document belongs according to the comparison result. A classification result display unit 617 for displaying a list of classified documents to be classified, and a plurality of appropriate categories are selected and indicated on the estimation result display unit when the estimation result by the category estimation unit is contrary to the user's intention. By correcting the importance value in the evaluation item table 618, It consists of learning means 619 Metropolitan to learn.
[0023]
The number of effective words in the document held in the effective word number holding unit 606 is given to the co-occurrence frequency calculation means 607, and the co-occurrence frequency calculation means 607 calculates the co-occurrence frequency of each set of effective words using the effective word number. The obtained co-occurrence frequency is stored in the co-occurrence frequency storage unit 608, and then provided to the effective word vector calculation unit 609. The effective word vector calculation means 609 obtains an effective word vector of each effective word using the co-occurrence frequency. Here, assuming that the co-occurrence probability of the valid word Ti and the valid word Tj is ci, j and the number of valid words is N, the valid word vector Ti of the valid word Ti is given by the following equation (1).
Ti = (ci, 1, ci, 2, ..., ci, N) (1)
It becomes.
[0024]
The co-occurrence probability ci, j is defined by the following equation (2).
[0025]
ci, j = (the number of documents including both Ti and Tj) / (the number of documents including Ti) (2)
The valid word vector obtained by the valid word vector calculation unit 609 is provided to the document vector calculation unit 611 after being stored in the valid word vector storage unit 610. The document vector calculation unit 611 obtains a document vector for each of the learning document and the classification target document by referring to the effective word vector. The document vector obtained for each of the learning document and the classification target document is a document vector holding unit. 612. The document vector of the learning document held in the document vector holding unit 612 is provided to the folder vector calculating unit 613, and the folder vector calculating unit 613 obtains a folder vector of each category using the document vector of the learning document. The obtained folder vector of each category is stored in the folder vector storage unit 614.
[0026]
The folder vector of each category held in the folder vector holding unit 614 is provided to the category estimating unit 615 together with the document vector of the classifying target document held in the document vector holding unit 612. The document vector is compared with the folder vector of each category, and the category to which the classification target document belongs is determined according to the comparison result. The determined category of the classification target document is stored in the classification result storage unit 616.
[0027]
Next, the processing procedure of the learning phase in the genre estimation device will be described with reference to FIG.
[0028]
First, in step 701, the learning document is subjected to morphological analysis, and words useful for classification are selected as effective words from the words included in the learning document. In step 702, the selected effective words included in each document are selected. Find the number of. Effective words are selected from common nouns, proper nouns, sa-variant nouns, and unknown words that appear in a particular category.
[0029]
Next, each effective word is weighted. The evaluation of the weight is performed from two viewpoints. That is, first, (1) the extent to which the effective word itself is effective for the act of classification. (2) How important the effective word indicates in each document.
[0030]
The weight w1 of (1) represents the degree of belonging to each category, and is based on the idea that an effective word having a higher degree of characterizing a specific category has a higher weight, and is calculated in the following manner. First, the ratio Pik of the document containing the valid word Wi in the learning documents belonging to the category Ck is obtained.
[0031]
Pik = (number of documents belonging to category Ck and including valid word Wi) / (number of documents belonging to category Ck)
However,
ΣPik (sum over all categories) = 1
Normalize so that
[0032]
W1 = 1−H (Wi), where H (Wi) is the entropy of Pik
Is defined.
[0033]
The weight w2 of (2) evaluates how the effective word is used in the target document and how it relates to the contents of the document. For example, "... In the United States, where the television is a mass media nation, politics is like a circus, and good students will not want to be politicians. , The word "circus" was only used as a metaphor for "politics" in the United States, and had little direct relevance to the text. .
[0034]
Therefore, if the word “circus” itself highly characterizes a specific category of “entertainment”, and if the effective word is given a high weight, the document vector will be pulled in the wrong direction. Therefore, it is necessary to evaluate the weight as in (2) in addition to (1) (the weight in (2) will be referred to as the document importance).
[0035]
Evaluation items are created in advance by focusing on (1) the appearance position of the valid word and (2) the linguistic role of the valid word, such as the case role and the modification type, as elements relevant to the importance in the document. In advance, a value of a weight given when an effective word satisfies the condition of each evaluation item is obtained by learning.
[0036]
First, the weight (1) will be described.
[0037]
It is considered that the appearance position of the effective word in the document has a strong correlation with its importance. For example, it is a well-known fact that in newspaper articles, important words appearing in the first paragraph that express meaning are apparent. Therefore, the entire document is equally divided into n blocks in units of sentences, and a weight for each block is obtained.
[0038]
Next, the weight (2) will be described.
[0039]
As a correlation between the linguistic role and the importance of the effective word, for example, "words used in the form of a subject such as" ... ga "," ... ha "are important" Is not possible. ' Therefore, the following evaluation items are prepared as linguistic roles, and the respective weights are obtained by learning.
Ga
""
"Ni" case
"He" case
"Ha" case
"Mo" case
Other nominative nomenclature
Nominal modifier
Modification of the noun
End of sentence
Sentence with reading point
End of sentence
In the present embodiment, as shown in FIG. 8, evaluation items such as □ valid word “is included in the first sentence of the paragraph”, □ valid word “ga” case, “wo” case, etc. "Case role", □ Effective words are "sentence nouns at the end of the sentence". The evaluation items include the position of the effective word in the document and the role of the dependency of the word. It is also possible to employ. FIG. 8 shows the initial state of the evaluation item table, and the values of the importance in each document are all “1.0”.
[0040]
Next, in step 703, the co-occurrence frequency of each set of valid words is obtained from the number of valid words, and in step 704, an effective word vector is calculated from the co-occurrence frequency. Then, an effective word is extracted from the learning document by referring to the effective word vector, and in a succeeding step 705, a document vector of the learning document is obtained by averaging the effective word vectors of the extracted effective words.
[0041]
Next, the process proceeds to step 706, where the average of the document vectors of all the learning documents belonging to each category is calculated, and the average is set as the representative vector of the category, and the present process ends.
[0042]
Next, a learning algorithm of the value of the in-document importance will be described with reference to FIG.
[0043]
First, the values of the in-document importance for all the evaluation items are initialized to one.
[0044]
Next, a counter variable n for counting the number of learning documents used for learning is initialized to "0" (step 901). Next, the learning document to which the correct classification category (= C *) is added is read, and the counter variable n is incremented by "1" (step 902). As an example of the read document, a document as shown in FIG. 10 which is classified into the category “incident” is assumed. Then, an effective word described in the effective word dictionary 604 is extracted from the read learning document, and an attribute table of the extracted effective words is created according to the evaluation items of the evaluation item table 618 (step 903). Here, FIG. 11 shows an example of valid words extracted from the document of FIG. 10 according to the valid word dictionary 604 and an attribute table relating to the valid words.
[0045]
Next, the importance of each effective word in the document is calculated based on the extracted importance table and the value of importance described in the evaluation item table 618 (step 904). Then, the degree of belonging to each category of the document is calculated using the calculated in-document importance, the degree of belonging of each valid word to the category held in the valid word dictionary 604, and the like, and the highest degree of belonging is calculated. The category is set as the classification result (= C) (step 905).
[0046]
Then, the classification result (C) is compared with the correct classification category C * acquired in step 902 to check whether or not the classification result C is correct (step 906). The value of the importance is corrected (step 907).
[0047]
Here, a method of updating the importance value will be described with reference to FIG. 8 and FIGS.
[0048]
Now, if the document in FIG. 10 is misclassified into the category "science", first, the effective words "Faculty of Engineering", "Laboratory", and "Chemical Experiment" which have a high degree of belonging to the category "Science", which is a misclassification. Focus on it. Since these valid words are considered to be the cause of misclassification, the value of the importance in the evaluation item table of FIG. 8 is corrected so that the importance in the document is reduced.
[0049]
That is, when attention is paid to “engineering department”, the attribute value of “whether or not in the paragraph head sentence” (evaluation item 1 in FIG. 8) in the attribute table of FIG. 11 relating to “engineering department” is “TRUE”. , “Case role” (evaluation item 2 in FIG. 8) is “union of“ no ””, so “TRUE” and “case role” of “whether or not in first sentence of paragraph” in evaluation item table **. Decreases the value of the importance of the “no” union ”by a small amount. When focusing on“ laboratory ”and“ chemical experiment ”, the same processing as“ engineering department ”is performed.
[0050]
Next, attention is paid to the valid words “fire” and “injury” that have a high degree of belonging to the correct classification category “incident”. In order to produce a correct classification result, the importance of these effective words in a document must be greatly evaluated.
[0051]
Therefore, when focusing on "fire", the attribute value of "whether or not in the first sentence of the paragraph" (evaluation item 1) in the attribute table relating to "fire" is "TRUE" and "case role" ( Since the evaluation item 2) is “ga” case, the importance level of “TRUE” in “whether or not in the first sentence of the paragraph” in the evaluation item table 618 and the importance level of “ga ka” in “case role” are evaluated. Increase the value by a small amount.When focusing on "injury", the same processing as "fire" is performed.
[0052]
After the importance value is updated in this way, the process proceeds to step 908, where the correct answer rate r of the classification for the past N (N ≦ n) learning documents is calculated. If it is determined in step 906 that the classification result is correct, the process of updating the value of importance in step 907 is skipped, and the process proceeds to step 908.
[0053]
Next, it is checked whether the correct answer rate r has exceeded a predetermined value Th or the number of learning documents used for learning has exceeded a predetermined value M (step 909). If none of the above is satisfied, the process returns to step 902, and the same processing is performed based on the next learning document.
[0054]
By performing such processing, the evaluation item table 618 in which the importance of each evaluation item is appropriately corrected is realized.
[0055]
In this way, specific importance values for evaluation items that are considered useful for evaluating importance in a document, such as the appearance position of a word, case role, modification type, etc., are stored for each category and are stored separately. Is obtained by learning using the learning document. That is, initially, the initial value of the importance of each evaluation item is appropriately given, and the learning documents are classified. If the classification result is different from the correct category, and an erroneous classification occurs, An effective word that has had a great influence on the misclassification is extracted, and the value of the importance assigned to the evaluation item of the importance in the document that applies to the case is corrected by a small amount. Such processing is performed on a large number of learning documents, and a value of the in-document importance that automatically maximizes the classification accuracy is automatically obtained.
[0056]
Next, referring to FIG. 12, a description will be given of a processing procedure in a classification phase for automatically classifying a classification target document whose category is actually unknown.
[0057]
In the classification phase, first, in step 1201, an effective word is extracted from the document to be classified by referring to the effective word vector obtained in step 704, and the effective word vector extracted in step 1202 (the effective word obtained in step 704). Vector), and a document vector of the document to be classified is obtained from the average of the vectors.
[0058]
Next, the process proceeds to step 1203, where the document vector of the document to be classified is compared with the folder vector obtained in the learning phase, the category to which the document to be classified belongs is determined according to the comparison result, and the process ends.
[0059]
If the digitized documents stored on the document management server are classified into categories by the method described above, the category can be similarly estimated for newly read paper documents.
(3) Date search processing
Returning to the flowchart of FIG. 2 again, the date search process will be described.
[0060]
First, a search is made from the morpheme sequence for a sequence that is considered to represent a date. At this time, if there is a line feed code or a plurality of spaces or tabs before and after the date portion, it is picked up as a portion in which only date-only data exists (step 203 in FIG. 2). If there is date data picked up, the date data is set as a creation date as a file attribute (step 204). In the case of the original image shown in FIG. 3, the line “(space or tab) May 30, 2001 (line feed)” exists in the third line counted from the first line. May 30, 2001, ** hours ** minutes ** seconds ". Here, since a portion corresponding to the hour, minute, and second could not be extracted by morphological analysis, a default time is assigned by the system.
[0061]
In step 203, if there is no date data picked up, a search is made from the divided morpheme strings for date data other than a single date row (step 205).
[0062]
Next, of the searched date data, those representing the present or future date are discarded, and only the past date data is left (steps 206 to 208). In the example of FIG. 3, if there is no third line regarded as a date line, “June 20 (Sun) 9:00 to 15:00” and “June 10 (Thu) )] Is extracted.
[0063]
After all the morpheme strings have been searched, heuristic estimation of the date on which the manuscript was actually created, or the date as close as possible, is performed from the extracted date data (step 209). At this time, the genre of the document estimated by the genre estimation processing is considered. For example, in a proposal, quotation, report, etc. created in an office, if multiple past dates appear, the date that goes back to the past is considered closer to the creation date. .
[0064]
On the other hand, news stories and news articles found in clippings of newspapers and magazines deal with past events or events that will occur in the future more than the moment they were written. Is determined from the contents and the date data extracted, etc., in the former case, the latest date of the extracted date data, and in the latter case, the oldest date is the date when the document was created. This data is assigned as a file attribute because it is concluded that the data is closest to the data.
[0065]
Also, when the creation date of the document is estimated from the date extracted in this way, the nature differs from the creation date given when the data is digitized data from the beginning created by keyboard input or the like, A flag is set as an identifier indicating that the information has uncertainty (step 210).
[0066]
If no date data exists, the file attribute assigned by the computer system is assigned as it is (step 211).
[0067]
<Other embodiments>
In the above-described embodiment, the genre of the original document is estimated based on the content of the text written in the document by using the electronic document group that has already been stored and automatically classified. Rather than classifying the documents, layout information and format patterns are registered in advance, and the format pattern of the document is recognized by an image recognition technique, and a genre is determined according to the format pattern.
[0068]
The timing of performing the genre estimation processing and the date data extraction processing on the document need not be as described in the embodiment, and the processing amount of the genre estimation processing and the The timing is adjusted to reduce the load on the system.
[0069]
【The invention's effect】
As described above, according to the present invention, in a document management server in which an already created digitized document and an OCR document in which a paper document is read by a scanner and character-recognized coexist and managed, When scanning and digitizing a document, search for the result of character recognition to find a character string that seems to contain the date the document was actually created, and assign it as a file attribute The present invention provides a paper / electronic file integrated document management system in which the operator can perform a search not on the date and time of scanning but on the date of actual creation or on the date close to the text, instead of the date and time of scanning. be able to.
[Brief description of the drawings]
FIG. 1 is a diagram showing an overall configuration of a system according to the present invention.
FIG. 2 is a flowchart showing a flow of processing in the present invention.
FIG. 3 is a diagram illustrating an example of a document image.
FIG. 4 is a diagram illustrating an example of a recognized character candidate.
FIG. 5 is a diagram showing an example of word candidates extracted by searching a dictionary.
FIG. 6 is a diagram showing a configuration of a genre estimation device.
FIG. 7 is a flowchart illustrating a processing procedure of a learning phase in the genre estimating apparatus.
FIG. 8 is a diagram showing an example of an evaluation item table.
FIG. 9 is a flowchart illustrating a learning algorithm of a degree of importance in a document.
FIG. 10 is a diagram illustrating an example of a classification target document.
FIG. 11 is a diagram illustrating an example of an attribute table.
FIG. 12 is a flowchart showing a processing procedure of a classification phase in the genre estimating apparatus.
[Explanation of symbols]
101 OCR
102 Morphological analyzer
103 Word Dictionary
104 Genre Estimation Device
105 Date data search device
106 Attribute Assignment Device
107 Document storage device

Claims (5)

A document management device that stores and manages document images and digitized documents read by a scanner or the like,
The character string in the document image is subjected to character recognition using a recognition dictionary, morphological analysis is performed using a word dictionary, and word output means for outputting a word as a recognition candidate is output by the word output means. An extracting device for extracting a portion representing a date and time from a word string; an estimating device for estimating a time when the document image was created from data on the date and time extracted by the extracting device; and an estimating device. And a providing unit for providing the estimated time as an attribute of the document image. Valid word extracting means for extracting characteristic words as valid words for automatically classifying documents into predetermined categories; and estimating a category to which the document belongs from the appearance degree of the valid words extracted by the valid word extracting means. 2. The document management apparatus according to claim 1, further comprising: a category estimating unit; and a category adjusting unit that changes a time estimated by the estimating unit according to the category estimated by the category estimating unit. 2. The apparatus according to claim 1, further comprising: a format discriminating unit for discriminating a format of the document image from layout information and the like, and a format adjusting unit for changing a time estimated by the estimating unit according to the format discriminated by the format discriminating unit. Document management device as described. 2. The document management apparatus according to claim 1, wherein, when the extraction unit fails in the extraction, the date and time at which the original image was read are added as attributes assigned by the assignment unit. The creation means extracted and attached from the text and the creation date as the date and time when the reading operation was performed can be distinguished and displayed as an attribute. 2. The document management apparatus according to claim 1, wherein a case where the document image is read as an attribute of the document image and a case where the date and time when the document image is read are provided as attributes of the document image can be distinguished.

Images