JP4241115B2 - Document filing device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a technique for digitizing paper documents.
[0002]
[Prior art]
A paper document is an excellent medium for transmitting and recording information, but requires a space for storing it. In addition, when information is recorded and stored in a paper document, when the paper document becomes necessary later, the target paper document must be searched from among many paper documents stored in a library. Therefore, the form of recording and storing in a paper document is not preferable from the viewpoint of improving business efficiency. Under such circumstances, paper documents are digitized and stored. Specifically, the user sets a paper document page by page in the scanner device, reads each page of the paper document with this scanner device, and stores it in an electronic file as image data.
[0003]
Such a method is inefficient when the number of pages of a paper document to be converted into an electronic file is large. For this reason, when the number of pages of a paper document is large, each page of the paper document is continuously used by using a scanner device having an automatic paper feeding mechanism (hereinafter simply referred to as ADF) called an auto document feeder (ADF). Automatic reading is widely performed.
[0004]
By the way, depending on the situation, a large number of paper documents to be digitized may be generated at one time. In such a case, if each paper document is set in the ADF one by one and read processing is performed, the paper document can be converted into an electronic file in document units. However, when this method is adopted, a paper document must be set in the ADF for each document and an operation for performing a reading process must be performed, which is cumbersome for the user. On the other hand, if all types of paper documents are collectively set in the ADF and read processing is performed, the plurality of types of paper documents can be collectively converted into an electronic file. However, when this method is adopted, image data of a plurality of types of paper documents are stored together in one electronic file, which is inconvenient for searching and browsing individual paper documents later.
Therefore, several technologies have been proposed in the past to meet the need for a technology that reads multiple types of paper documents at once and converts the image data of paper documents into electronic files for each document. Has been.
[0005]
For example, Patent Document 1 discloses the following electronic filing device. First, when there are a plurality of documents to be digitized, the user of the apparatus inserts a specific document separator document between the plurality of documents in advance. Then, an original composed of a plurality of documents into which the document separation original is inserted is set on the ADF, and the apparatus is caused to perform image reading processing. The apparatus recognizes image data corresponding to a document separation document from page image data of all pages obtained by the image reading process, and converts the page image data into an electronic file for each document.
[0006]
[Patent Document 1]
JP-A-1-162474
[0007]
Patent Document 2 discloses the following electronic filing device. That is, when there are a plurality of documents to be converted into electronic files, the user of the apparatus adds a document delimiter mark in advance to the reverse side of the page that becomes a document delimiter in each document. In this apparatus, both the front and back sides of each page of a document composed of a plurality of documents are read by an image scanner, the document separation mark is recognized, and the page image data of each document is converted into an electronic file for each document.
[0008]
[Patent Document 2]
JP-A-7-287747
[0009]
Patent Document 3 discloses the following electronic filing device. That is, the apparatus cuts out a character recognition area portion set in advance from image data read by an image scanner, performs character recognition processing, and determines a document break based on the character recognition result.
[0010]
[Patent Document 3]
Japanese Patent Laid-Open No. 10-21380
[0011]
Further, there is a method in which a plurality of documents are collectively read by an image scanner having an ADF, a feature amount of the read document image is calculated, and a document unit break is determined based on the feature amount. According to this method, the document separation can be determined without inserting a document separation document between the documents in advance or modifying the page that becomes the document separation, so that the burden on the user is greatly reduced. be able to. An example of such an apparatus is disclosed in Patent Document 4 below.
[0012]
[Patent Document 4]
Japanese Patent Laid-Open No. 2002-24258
[0013]
[Problems to be solved by the invention]
In any of the above prior arts, an ADF is used, but there is an upper limit on the number of documents that can be set in the ADF at one time. For this reason, there were the following problems.
[0014]
First, if the manuscript to be processed is composed of manuscripts of a plurality of types of documents (hereinafter referred to as document manuscripts) and is in large quantities, it is considered convenient for the user if the following work mode can be taken.
That is, the user takes out an appropriate amount of original bundles that can be set on the ADF from the stacked original bundles, sets the original bundles on the ADF, and causes the scanner device to read the original bundles. Hereinafter, this series of operations is referred to as a scan job. If such a scan job is repeated until there are no more original bundles, the image reading process for all the originals can be completed.
However, in such a work mode, for example, when a document bundle is set in the ADF and the Nth scan job is performed, the document bundle ends with a page in the middle of the document. Can happen. In this case, the remaining pages of the document are processed in the (N + 1) th scan job.
[0015]
In such a case, when the scanned image data is converted into an electronic file for each scan job, the image data acquired from the same document corresponds to the first half and the second half of the document. Are stored in separate electronic files as if they were image data of separate documents. This hinders the creation of electronic files in document units.
In order to avoid this, it is necessary for the user to visually check the separation of document originals before setting a bundle of originals so that processing of a single document original is not performed across two scan jobs. . However, such work is troublesome for the user.
[0016]
Furthermore, when the total number of pages of one document exceeds the ADF processing upper limit, there is a problem that a single document must be processed across a plurality of scan jobs. For example, when the number of originals that can be set in the ADF is 100 pages, a document original having a total number of 150 pages is processed.
[0017]
The present invention has been made in view of the circumstances described above, and processing of a single document original is performed across a plurality of scan jobs, and image data of the same document is divided and stored in two electronic files. It is an object of the present invention to provide a method and apparatus for document filing that can detect such a situation.
[0018]
[Means for Solving the Problems]
  In one aspect, the present invention provides a document filing apparatus including a control unit, a storage unit, and an original reading unit for reading an original, wherein the original reading unit represents one or more pages representing individual pages constituting the document. A first step of acquiring a plurality of document image data sets made up of image data and storing them in the storage unit; and the control unit refers to the storage unit and a first document image in the plurality of document image data sets. For each document image data included in the data set, a feature amount vector is extracted from all page image data in the document image data, and each feature amount vector extracted from each page image data forms in a feature amount space. And determining from the page image data of the first page in the second document image data set in the plurality of document image data sets Whether or not the first document image data set and the second document image data set are obtained from the same document depending on whether or not the feature quantity vector belongs to the cluster region. And a second method for determining whether or not a document is divided.
  In a preferred aspect, each of the plurality of document image data sets includes document image data read from one original bundle in one scan job by the original reading unit, and the first document image data set and The second document image data set is composed of document image data read from two original bundles in a continuous scan job.
  In another aspect, the present invention provides a document input unit that acquires a plurality of sets of document image data composed of one or a plurality of page image data representing individual pages constituting a document, and a first of the plurality of document image data sets. For each document image data included in one document image data set, feature quantity vectors are extracted from all page image data in the document image data, and each feature quantity vector extracted from each page image data is stored in the feature quantity space. A cluster area to be formed is determined, and a feature quantity vector is extracted from page image data of the first page in the second document image data set in the plurality of document image data sets, and the feature quantity vector belongs to the cluster area Whether the first document image data set and the second document image data set are the same. Providing a document filing apparatus including a determining document dividing determination section whether or not that is al acquired.
  In a preferred aspect, the document input unit includes an ADF that takes out the paper one page at a time from the set of originals, and a scanner device that reads the original from the paper sequentially taken out by the ADF and outputs page image data. A set of document image data is constituted by page image data obtained from one original bundle set in the.
  In another preferable aspect, the document segmentation determination unit treats two document image data sets obtained sequentially in time as the first document image data set and the second document image data set.
[0019]
In the following description, image data for one page of a document is referred to as “page image data”. Further, it is a set of page image data composed of image data of each page of a document, and the image data for one document is called “document image data”.
[0020]
DETAILED DESCRIPTION OF THE INVENTION
<A. First Embodiment>
<1. Configuration>
FIG. 1 is a schematic diagram of a document filing apparatus 1 according to the first embodiment of the present invention. As shown in the figure, the document filing apparatus 1 according to the present embodiment includes a document input unit 101 having an ADF, a document separator 102, a document storage unit 103, a document segmentation determination unit 104, and a feature amount extraction unit. 105, a similarity evaluation unit 106, a divided document connection unit 107, a user interface 108, and a document output unit 109.
[0021]
The document input unit 101 includes, for example, a scanner device having an ADF and its control unit. In the document input unit 101, when a document is set on the ADF by the user, the document is sequentially sent to the scanner device page by page and image reading processing is performed. Through this image reading process, page image data of the document is acquired, and is collected into one electronic file and stored in the document storage unit 103. In the present embodiment, the page image data acquired in one job scan as described above is collected into one electronic file and stored in the document storage unit 103.
[0022]
The document delimiter 102 includes a CPU, an image processor, a RAM, and a ROM. The document delimiter 102 reads each electronic file stored in the document storage unit 103 by the document input unit 101, and determines a delimiter position between documents in the page image data in the electronic file for each electronic file. Then, the document delimiter 102 executes predetermined document delimiter processing to delimit the page image data in the electronic file at the determined delimiter position, and the page image data divided in document units is stored in the document storage unit 103. To accumulate. Here, various existing methods can be used as the document separation determination method and document separation processing, and the method described in “Prior Art” may be used. As an example, a method is conceivable in which character recognition is performed on the read page image data, a document break position is specified, and the document is stored in the document storage unit 103 separately in document units.
[0023]
The document storage unit 103 includes a large-capacity storage device such as a hard disk drive or a DVD-RAM (Digital Versatile Disc-Random Access Memory) drive, and the page image data input by the document input unit 101 and the document delimiter unit 102 It is used as means for storing page image data divided into units.
[0024]
The document segmentation determination unit 104 includes a feature amount extraction unit 105 and a similarity evaluation unit 106, and whether or not a single document is scanned across a plurality of scan jobs, that is, a single document It is determined whether the page image data is divided into two electronic files and stored in the document storage unit 103.
[0025]
The feature amount extraction unit 105 is a means for extracting feature amounts from page image data. Further, the similarity evaluation unit 106 uses the feature amount extracted by the feature amount extraction unit 105, and page image data obtained by the Nth scan job and page image data obtained by the N + 1th scan job, Determine the similarity of. Here, the page image data used as a comparison target may be page image data for one page or image data for a plurality of pages. Details will be described later.
[0026]
When the document segmentation determination unit 104 determines that a single document has been segmented and read, the segmented document concatenation unit 107 performs concatenation / integration processing on the page image data of the document to obtain a single document image data Reconfigure to
[0027]
The user interface unit 108 is an input device such as a keyboard and a mouse, and inputs a document input instruction and a document output instruction from the user.
[0028]
The document output unit 109 outputs the input page image data to a predetermined place by a predetermined method in accordance with an instruction from the user. In the present embodiment, the following is assumed as the document output unit 109. That is, a display unit having a CRT (Cathode Ray Tube) or liquid crystal panel and its control hand throw, a printing unit having a printer device and a control unit, a read / write device such as a magnetic disk and a memory card, and its control unit, etc. A data transfer apparatus that exchanges data via a storage unit or a network. For example, the input page image data may be output to a CRT, the data may be edited in accordance with an instruction from the user, converted into an HTML (Hyper Text Markup Language) format, and stored in a storage medium.
[0029]
<2. Operation>
In the following description, it is assumed that a paper document consisting of a plurality of pages is divided into M bundles, and these divided bundles are processed by M scan jobs. Further, it is assumed that the user has already completed N scan jobs and will perform the (N + 1) th scan job from now on. However, M and N are natural numbers, and M> N.
[0030]
The user sets a paper document consisting of one or more pages in the ADF as the (N + 1) th scan job. At this time, the paper document may be a single document or a partial page of a single document. Alternatively, it may be composed of a plurality of documents. Further, when setting a paper document, the user does not need to be aware of document separation. The second half of a certain document may be included at the beginning of a paper document set by the user, or the first half of a certain document may be included at the end of the paper document. However, when the document separation unit 102 uses a technique for detecting the document separation position based on the document separation document, predetermined processing such as inserting the document separation document into the document separation position in advance is performed. Shall be kept. It goes without saying that the number of paper originals set in the ADF must not exceed the upper limit of the number of pages that can be processed at one time by the ADF.
[0031]
The paper document set in the ADF is sent to the scanner device page by page, and page image data corresponding to the number of pages of the set paper document is generated. The page image data is associated with the (N + 1) th scan job. And stored in the document storage unit 103 as an electronic file. For example, when 100 paper originals are set at a time and all sides are scanned, page image data for 200 pages is input.
[0032]
In order to prevent the explanation from becoming complicated, in the following explanation, the document d_nWhen (m) is described, it represents the mth read document in the nth scan job. And d_{P, K}When (L) is described, it is assumed that the Lth scan job represents a K page paper manuscript from the beginning of the Pth scanned document, and the page image data of this page is represented by D_{P, K}(L). Thus, for example, the second scanned document in the Nth scan job is d₂(N), and the page image data relating to the document is three page image data D_2,1(N), D_2,2(N) and D_2,3(N) (see FIG. 2).
[0033]
The input page image data is stored in the document storage unit 103 after the document separation unit 102 determines the document separation position and is separated into document units. As a method of storing separately for each document, a file folder is prepared for each document, and each page image data related to one document is numbered in the order of input, and the file name is used. There is a method to store in the file folder. Specifically, document d_mA folder name “document image data (job number, document n)” is assigned to the folder in which (n) is to be stored. For example, document d₂The folder name of the folder for storing the document image data according to (N) is “document image data (job N, document 2)”, and the page includes three page image data D_2,1(N), D_2,2(N) and D_2,3(N) is stored. Here, page image data D_{P, K}It is assumed that a data file name “page image data (job N, document M, page P)” is assigned to (L). For example, page image data D_2,1The file name of (N) is “page image data (job N, document 2, page 1)”.
[0034]
Next, as shown in FIG. 3, the document segmentation determination unit 104 performs a document d that is the last document in the Nth scan job._ThreeDocument d, which is the first document in the (N) and (N + 1) th scan jobs₁It is determined whether (N + 1) is the same document.
[0035]
Specifically, first, the feature amount extraction unit 105 performs document d._ThreePage image data D relating to the last page of (N)_3,3(N) and document d₁Page image data D relating to the first page of (N + 1)_1,1A feature amount is extracted from each of (N + 1).
More specifically, a feature vector representing the feature of page image data for one page is determined. The component of this vector is obtained as follows. That is, first, page image data D_3,3From (N), for example, quantities indicating the characteristics of the page image data such as the color tone, aspect ratio, luminance and color distribution state, edge distribution state, and flat area distribution state of the entire image are extracted. This extraction process will be described more specifically with reference to FIG. First, an area of one page image data is divided into X in the vertical direction and Y in the horizontal direction. That is, one page image data is divided into X × Y areas.
[0036]
Next, for each region, image analysis processing is performed to calculate a feature amount of the region. As described above, various parameters can be adopted as the feature amount, but here, the gradation is adopted as an example. More specifically, first, R, G, and B components of the color of one region are obtained. The R, G, and B components are converted into three quantities representing color components in the L * a * b space (characteristic color space). These three quantities are used as feature quantities corresponding to one area. As described above, (X × Y × 3) feature amounts are extracted from the page image data for one page.
In the left part of FIG. 4, the case of X = 6 and Y = 4 is shown as an example, and a total of 72 feature values are calculated from this page image data (see the center of FIG. 4). That is, one 72-dimensional vector having 72 components is generated (see the right side of FIG. 4). Similar processing is performed for page image data D._1,1Repeat for (N + 1).
[0037]
Next, the similarity evaluation unit 106 calculates the distance between the two vectors thus obtained. Here, the Euclidean distance or Mahalanobis distance is used as the distance. As shown in FIG. 5, when the calculated distance is smaller than a predetermined reference value, page image data D_3,3(N) and page image data D_1,1It is similar to (N + 1) and is determined to belong to the same document. In other words, document d_Three(N) and document d₁(N + 1) is determined to be the first half and the second half of the same document (see the left side of FIG. 5). If the calculated distance is greater than a predetermined value, page image data D_3,3(N) and page image data D_1,1(N + 1) is dissimilar to and is determined to belong to a different document. In other words, document d_Three(N) and document d₁It is determined that the document is different from (N + 1) (see the right side of FIG. 5).
[0038]
In the document segmentation determination unit 104, the document d_Three(N) and document d₁If it is determined that (N + 1) is the same document, the divided document concatenation unit 107 determines that the document d_Three(N) and document d₁A process of linking (N + 1) is performed. Specifically, page image data D_3,1(N), D_3,2(N) and D_3,3(N) and page image data D_1,1(N + 1), D_1,2(N + 1), and D_1,3Combine (N + 1).
More specifically, first, page image data D stored in the folder “document image data (job N + 1, document 1)”._1,1(N + 1) D_1,2(N + 1) D_1,3(N + 1) is transferred to the folder “document image data (job N, document 3)”. Next, the data file name of the migrated data is changed. That is, the file name of the migrated data is the page image data D_3,3Rename the file so that it follows (N). Specifically, the file name “page image data (job N + 1, document 1, page 1)” is “page image data (job N, document 3, page 4)” and page image data (job N + 1, document 1, page 1). "Page 2)" is "page image data (job N, document 3, page 5) and page image data (job N + 1, document 1, page 2)" is "page image data (job N, document 3, page 6)". And change.
[0039]
The above processing is performed for all continuous scan jobs. That is, when performing M scan jobs in total, if the first scan job is called scan job 1, the continuous scan jobs are scan job 1 and scan job 2, scan job 2 and scan job 3,. There are a total of M scan jobs M-1 and M, and scan job M and scan job 1. Therefore, the document segmentation determination process described above is performed a total of M times for all the document image data related to successive scan jobs in order from the first head scan job.
In this way, since only the page image data of all pages constituting one document is stored in each folder, all the page image data divided in the M bundle of paper originals are correctly combined and combined. Can be stored.
[0040]
When the document image data is stored using an image file format such as Tiff that can hold a plurality of page images, the document image data is stored in the folder “document image data (job N, document M)” after performing the above-described processing. All the existing page image data are combined to generate one file. The file name may be the same as the folder name, for example.
[0041]
Note that the document d which is the first document in the (N + 1) th scan job₁(N + 1) and the document d which is the last document in the Nth scan job_ThreeWhen performing the division determination with (N), page image data stored in the folder “document image data (job N, document M)” (M is an arbitrary integer) may not exist. This is a case where one document is scanned over three or more. In this case, since the page image data obtained by the Nth scan job has already been combined with the page image data related to the N−1th and lower scan jobs, the folder is empty.
As described above, when the page image data to be determined for division does not exist in the folder related to the immediately preceding scan job, the page image data related to the previous scan job is combined. That is, until the page image data to be combined is found, the search is sequentially performed up to the folder with the lower folder number, and if it is found, the document is determined to be divided. If it is determined that the combination is necessary, the same combination process as described above is performed.
[0042]
If the scan job number of the generated page image data is unnecessary for the user, the generated folders may be arranged in the order in which the scan job is executed, and a new number may be assigned to each document. In this way, since a serial number for distinguishing stored documents is assigned to each folder, a simpler arrangement of document image data is realized without using a scan job number.
[0043]
B. Modified example
<1. Features>
In the embodiment described above, information such as the color tone, aspect ratio, luminance and color distribution state, edge distribution state, or flat area distribution state of the entire image is extracted as characterizing the page image data. Such information is very effective when an image occupies all or most of the original document such as a presentation material. However, in the case of a manuscript consisting mostly of characters, such as a contract, it may not be very effective. Therefore, in the present modification, instead of extracting information such as color tone and brightness from the page image data, information indicating the characteristics of the document format such as character height, character spacing, line spacing, vertical writing / horizontal writing (hereinafter, referred to as “character writing”). The document format feature value is extracted, and it is determined whether the two documents are the same based on the document format feature value.
[0044]
The page image data comparison method according to the present embodiment is effective for determining whether two pages of document images belong to the same document or different documents. This is because, since it is common for a document to have a uniform appearance, page image data with similar document format characteristics as described above can be assumed to belong to the same document. . Conversely, page images belonging to different documents rarely have the same format characteristics unless they are intentionally created in the same format. Therefore, in the case of a document composed almost of characters, the similarity of the page image data can be accurately determined.
[0045]
Various conventional techniques can be applied to obtain format characteristics such as character height, character spacing, and line spacing from a document image input from a scanner or the like. An example of such a prior art is disclosed in paragraph Nos. 0009 to 0012 of JP-A-5-108793. In the technique, b is a line spacing and d is a character spacing on the premise of horizontal writing. However, since the line spacing is usually larger than the character spacing, comparing b and d, the larger one is the line spacing, the smaller one is the character spacing, horizontal writing when b is larger, and vertical when d is larger. Whether writing or vertical / horizontal writing may be determined.
[0046]
In general, even for pages belonging to the same document, the font size, line spacing, and the like may be slightly different depending on the parts constituting the document, such as the title line, summary section, and body. For example, the title character is set to a font larger than the text. Therefore, even if the format features described above are extracted, the values differ depending on the parts constituting the document and are generally not consistent throughout the page. If the value of the format feature is not consistent, it is difficult to determine whether or not the document image of each page belongs to the same document using the format feature amount. This problem can be dealt with as follows.
[0047]
First, similarly to the method described with reference to FIG. 4, the input page image data is divided into a plurality of regions. Next, a format feature amount including information such as character height, character spacing, line spacing, vertical writing and horizontal writing is calculated for each of the divided areas. For example, the values of character height “10 pt”, character spacing “15 pt”, line spacing “10.5 pt”, vertical writing / horizontal writing “0” are obtained. Here, “0” is associated with vertical writing, and “1” is associated with horizontal writing.
[0048]
Next, a value that appears most frequently among the calculated feature values of all the regions is determined, and this is used as the feature value of the page. For example, if the height of a character is “10 pt” in 20 areas out of a total of 24 areas, “12 pt” in 3 areas, and “16 pt” in one area, the page concerned “10 pt” is calculated as the feature amount related to the height of the character. The same applies to character spacing, line spacing, vertical writing, and horizontal writing.
[0049]
Generally speaking, the ratio of titles and headings to the entire page is small compared to the body that characterizes the document. Therefore, the most frequently occurring value can be regarded as an amount representing the feature of the text. As a result, even if the document has a title and a heading, the feature amount can be accurately obtained from the page image data.
[0050]
In this way, a feature vector (hereinafter referred to as a format feature vector) representing the document format feature of the page is obtained from page image data for one page. An example of the format feature vector f is shown below.
[0051]
[Expression 1]
f = (character height, character spacing, line spacing, [vertical writing: 0, horizontal writing: 1])
[0052]
In this case, it is a four-dimensional vector having four components.
[0053]
Next, the similarity evaluation unit 106 obtains the distance L between the format feature vectors obtained from each of the two page image data whose similarity should be compared. If the feature vectors are fa and fb, respectively, the distance L can be expressed as follows.
[0054]
[Expression 2]
L = ‖fa-fb‖
[0055]
In the above equation, ‖v‖ represents the norm of the vector v. If L is smaller than a predetermined value, it is determined that the two page image data are similar and belong to the same document. When L is larger than a predetermined value, it is determined that the two page image data have no similarity and belong to different documents.
[0056]
<2. Comparison target image data>
In the above embodiment, the document d, which is the last document in the Nth scan job._LAST(N) and document d which is the first document in the (N + 1) th scan job₁In determining whether to link (N + 1), the document d_ThreePage image data D relating to the last page of (N)_3,3(N) and document d₁Page image data D relating to the first page of (N + 1)_1,1(N + 1) was compared.
However, the document d is not limited to this._Three(N) page image data relating to all pages and document d₁You may compare the page image data concerning all the pages of (N + 1). Specifically, as shown in FIG._ThreeFeature values are extracted from each of the page image data related to the page (N) by the same method as in the above embodiment, and feature vector vectors for the number of pages are generated. Next, an average vector of these vectors is obtained, and this is expressed as document d._ThreeLet (N) be the feature vector. This process is document d₁Similar processing is performed for (N + 1). Then, as shown in FIG. 7, by calculating the distance between the two feature vectors thus obtained, the document d_Three(N) and document d₁The similarity of (N + 1) is determined. The feature amount here may be an image data feature amount or a document format feature amount.
[0057]
<3. Comparative judgment method>
In the above embodiment, the similarity of page image data is determined by calculating the distance between vectors. However, the present invention is not limited to this, and the similarity of image data may be determined using a clustering method. Specifically, as shown in FIG. 8, for the Nth scan job and the (N + 1) th scan job, feature amounts are extracted for page image data of all read pages, as in the above embodiment, and a vector space Map within.
Next, an appropriate clustering method is applied to the set of feature values mapped in the vector space to divide it into clusters. Various methods exist as the clustering method, and an example is disclosed in Japanese Patent Application Laid-Open No. 2001-256244.
And document d_LASTThe page image data of the last page of (N) and the document d₁Page image data D of the first page of (N + 1)_1,1If one point in the vector space indicating the (N + 1) feature quantity belongs to the same cluster, it is determined that the two documents belong to the same document.
[0058]
<4. Features and Extraction Method>
The feature amount to be extracted from the small area may be one of the color tone, aspect ratio, luminance and color distribution state, edge distribution state, and flat area distribution state of the entire image, or may be plural. Good. For example, when the area of the page image data is divided into 24, the tone and edge state may be extracted from one small area, and the feature amount may be expressed as a 144-dimensional vector. Further, the extracted feature amount may be reduced by using a method such as principal component analysis without damaging the features of the page image data as much as possible. As a result, the amount of calculation to be performed by the document division determination unit is reduced, and the time required for the calculation is reduced. In other words, the number of parameters to be extracted is arbitrary. The number of divided areas is not limited.
[0059]
The document filing apparatus 1 may analyze the obtained page image data and determine the extraction method and the type and number of parameters to be extracted based on the analysis result. In the determination, it is desirable to compare and consider the accuracy of similarity determination, the amount of calculation, and the determination speed.
Alternatively, the document filing device 1 may allow the user to specify the determination accuracy and the processing time at the start of scanning or at the time of determination of division. In this case, the document image filing device 1 determines the extraction method, the type and number of extraction parameters, and the like based on designation from the user.
[0060]
<5. Scanner Device>
In the above embodiment, the document input unit 101 has been described on the assumption that the document input unit 101 includes a single scanner device. However, it is also possible to configure the document input unit 101 from a plurality of scanner devices and process a plurality of scan jobs in parallel.
For example, consider the case of using 10 scanner devices. If a total of 1000 originals are processed, the user first divides the original into 100 original bundles. Then, each document bundle is sequentially set in the scanner apparatus in the order of the assigned number, and each document bundle is read.
In this case, the problem is how to define adjacent scan jobs (Nth scan job and N + 1th scan job), which can be solved by using the following method.
[0061]
That is, as a first method, a number is assigned in advance to uniquely identify a plurality of scanners. For example, consider the case of using 10 scanner devices. If a total of 1000 originals are processed, the user first divides the original into 100 original bundles. Then, each document bundle is sequentially set on the ADF of the scanner apparatus in the order of the assigned number.
Specifically, assuming that there are K scanners in total, the document linking process is called scanner 1 and scanner 2, scanner 2 and scanner 3,..., Scanner K-1 and scanner K, scanner K and scanner 1. It only occurs in combination. Therefore, the document segmentation determination unit 104 only needs to determine segmentation for scan jobs of these combinations.
[0062]
As a second method, first, the document filing apparatus 1 is provided with an apparatus for acquiring the time when each of a plurality of scanners starts reading a document. Then, the document division determination unit 104 determines the division only for the scan job in the scanner device whose scan start time is adjacent. This is because the reading process is started in the order in which the user sets a paper document in each scanner device.
[0063]
【The invention's effect】
Even when the same document document is read across two or more scan jobs, the user does not have to manually combine the page image data divided and stored.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an overall configuration of a document filing apparatus 1 according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating a document input in one scan job.
FIG. 3 is a diagram illustrating pages to be compared.
FIG. 4 is a diagram illustrating a method of extracting page image data feature values.
FIG. 5 is a diagram illustrating a method of determining similarity / dissimilarity of page image data.
FIG. 6 is a diagram showing pages to be compared in one modified example.
FIG. 7 is a diagram illustrating a feature amount comparison method in the modification.
FIG. 8 is a diagram showing a method of determining similarity / dissimilarity of page image data in another modified example.
[Explanation of symbols]
1 ... Document filing device
101 ... Document input part
102 ... Document separator
103 ... Document storage unit
104... Document division determination unit
105... Feature amount extraction unit
106: Similarity evaluation unit
107 ... Document division connection
108 ... User interface
109 ... Document output section

Claims (5)

In a document filing apparatus including a control unit, a storage unit, and a document reading unit that reads a document,
A first step in which the document reading unit acquires a plurality of document image data sets composed of one or a plurality of page image data representing individual pages constituting a document and stores them in the storage unit ;
Wherein the control unit, referring to the storage unit, for each document image data included in the first document image data set in the previous SL plurality of document image data set, features from all the page image data in the document image data A vector is extracted to determine a cluster area formed in each feature quantity space by each feature quantity vector extracted from each page image data, and the first page in the second document image data set in the plurality of document image data sets The feature vector is extracted from the page image data, and the first document image data set and the second document image data set are obtained from the same document depending on whether or not the feature vector belongs to the cluster area. And a second step of determining whether or not the document has been processed. Each of the plurality of document image data sets is constituted by document image data read from one original bundle in one scan job by the original reading unit , and the first document image data set and the second document image data set. 2. The document fragmentation determination method according to claim 1, wherein the document image data set is composed of document image data read from two original bundles in a continuous scan job. A document input unit for obtaining a plurality of sets of document image data composed of one or a plurality of page image data representing individual pages constituting the document;
For each document image data included in the first document image data set in the plurality of document image data sets, feature amount vectors are extracted from all page image data in the document image data, and each extracted from each page image data Determining a cluster region formed by the feature amount vector in the feature amount space, extracting a feature amount vector from page image data of a first page in the second document image data set in the plurality of document image data sets, and Document fragmentation for determining whether the first document image data set and the second document image data set are obtained from the same document depending on whether a feature vector belongs to the cluster area A document filing device having a determination unit. The document input unit includes an ADF that takes out paper one page at a time from a set of originals set, and a scanner device that reads the originals from the paper sequentially taken out by the ADF and outputs page image data, and is set in the ADF The document filing apparatus according to claim 3, wherein one document image data set is constituted by page image data obtained from one document bundle.   The document fragmentation determination unit treats two document image data sets obtained successively in time as the first document image data set and the second document image data set. Document filing device described in 1.

Images