KR20180077573A - Document security method - Google Patents

Abstract

The present invention relates to a document security method. The me document security method comprises the steps of: distributing and storing an imaged document in a plurality of DB servers after puzzling the document, and collecting the distributed and stored images in order to obtain a document image; and restoring a puzzled image. It is possible to prevent a leakage of document information due to hacking of an individual DB server. Also, even when all DB servers are hacked, document information cannot be obtained. In addition, puzzling information is simplified to the number of progression of seed values of a pseudo random number generator, and puzzling is probabilistically appropriately performed. Accordingly, it is possible to secure security due to puzzling without excessive or insufficient puzzling and search document contents. An original document is provided by an image-document conversion, and thus utilization is enhanced. A study related to the present invention was supported by ″The Cross-Ministry Giga KOREA Project″ grant from the Ministry of Science, ICT and Future Planning, Korea ([GK16P0100, Development of Tele-Experience Service SW Platform based on Giga Media]).

Description

Document security method {DOCUMENT SECURITY METHOD}

The present invention divides an imaged document into a plurality of image pieces and puzzles them, distributes them to a plurality of DB servers, acquires a distributed image and obtains a document image, and restores the puzzle image And a document security method.

The research related to the present invention was carried out with support from Giga KOREA business [GK16P0100, Giga Media based tele-experience service SW platform technology development] of the future Creation Science Department. (This work was supported by 'The Cross-Ministry Giga KOREA Project 'grant from the Ministry of Science, ICT and Future Planning, Korea [GK16P0100, Development of Tele-Experience Service SW Platform based on Giga Media]).

Existing technologies that distribute and store electronic documents related to payment information, contract information, personal information, genetic information, hospital treatment information, management information, and confidential information are divided into a plurality of DB servers, The information is stored in a separate DB server. Thus, even if individual DB servers are hacked, the original information can not be obtained completely.

When information is viewed or downloaded, information is collected from each DB server according to the distributed storage information and restored.

Also, in order to enhance the security, the information is encrypted and distributed to a plurality of DB servers.

However, in order to make the decryption difficult, the encryption process must be complicated and difficult, and the encryption process is performed separately from the distributed storage method, so that the complexity can not be reduced as it is dualized.

On the other hand, there is a technique of dividing and distributing an image obtained by scanning or graphicizing a document as described in Patent Document 10-2001-0044248.

This technique is useful for scanning paper documents to be distributed and stored after image formation, converting them into image format, and then archiving them.

However, although the images stored in the individual DB servers are included only in some image fragments, they are vulnerable to security because they contain information that can infer a complete original image. In order to solve this vulnerability problem, It is difficult to appropriately divide the number of divided groups according to the shape and the number of divided groups to be considerably large.

KR 10-0749428 B1 2007.08.08. KR 10-2001-0044248 A 2001.06.05.

Accordingly, the present invention can prevent the analogy of the original image from being encrypted by encrypting an image of a document, distributing the encrypted image, distributing and storing the encrypted image, and reducing the complexity by linking the image encryption process to the distributed storage process, The present invention provides a document security method that can reduce the complexity of image restoration.

In order to achieve the above object, the present invention provides a document registration system comprising: a document registration unit (10) for dividing a document into a plurality of DB servers (3) and storing document information for collecting distributed stored documents in a document information DB server (4) A document security method performed by a document restoring unit (20) that collects and restores distributed stored documents according to document information when a document is requested, the method comprising: a document registration step (S100) performed by the document registration unit (10); And a document restoring step (S200) performed by the document restoring unit 20.

The document registration step S100 includes a puzzling step S110 of dividing an image created as a graphical document or a graphic document into a plurality of pieces and changing the arrangement of image pieces to obtain a puzzled image, (S130) for dividing image pieces of the puzzled image into a plurality of groups and distributing the image pieces in a plurality of DB servers (3) , And a document information DB step (S140) of registering the distributed storage information and attribute information in the document information DB server 4 as document information.

The document restoring step S200 includes a retrieving step S210 of obtaining corresponding puzzling information and distributed storing information based on the attribute information of the corresponding document when a document is requested to be provided, (S220) for obtaining combined images after collecting, and an inverse puzzling step (S230) for obtaining an image before puzzling by rearranging in reverse to the arrangement change when puzzling based on the puzzling information.

The document decompressing step S200 may include a step S240 of obtaining a document before graphing by recognizing text if the image obtained in the reverse puzzling step S230 is an image obtained by converting the document into a graphic form.

In the attribute extracting step S120, the text in the document is extracted and included in the attribute information. In the searching step S210, the text included in the attribute information is searched to specify the attribute information, and the puzzling information corresponding to the specific attribute information And distributed storage information.

In the distributed storage step (S110), image fragments on the puzzled image are left unchanged for each group, and a sparse image in which the remaining area is left empty is generated and stored in different DB servers (3) S220 overlap the scarcity images taken by each DB server 3.

In the distributed storage step (S110), the image fragments on the puzzled image are left unchanged for each group, and a sparse image is created by filling the remaining area with an arbitrary image, and stored in different DB servers (3) In the collecting step S220, the scarcity image taken by each DB server 4 is deleted after the arbitrary image is deleted.

In the puzzling step S110, a division number of the image is numbered, and a random number sequence of two times the shuffling number is generated by a pseudo random number generator in which a random number sequence is determined according to a seed And generating a puzzled image by repeating a shuffling process of taking two images in the order of the random number sequence and exchanging the image pieces of the two random number-assigned split positions, and the puzzle information includes a seed value and a shuffle And the inverse puzzling step S230 generates a random number sequence of two times the shuffling number according to the seed value, and then takes two pieces of the random number sequence in the reverse order of the random number sequence, The image before puzzling is obtained by repeating the shuffling process of exchanging the image pieces.

The shuffling count is a convergence rate to a stationary distribution in which one specific image fragment can be stochastically present at a uniform probability regardless of the dividing position as the shuffling process is repeated, It is determined as a function of the number of image fragments.

The degree of convergence is a value obtained by dividing the maximum deviation of the probability by the dividing position by the dividing position probability at the time of the normal distribution.

The shuffling count k may be calculated according to the number N of image pieces and the degree of convergence?

Respectively.

The present invention as described above divides an image into a plurality of image pieces, rearranges and puzzles the image, and then distributes a plurality of image pieces. Therefore, not only a leakage of individual storage images but also a complete restoration of original images .

In addition, the present invention improves the utilization of the restored image by image-document conversion.

In addition, the present invention minimizes the complexity in restoration by adding image puzzling process by making distributed images to be scarce images.

In addition, the present invention can reduce the complexity of the document management and restoration even if image puzzling is added, since the puzzle can be reversed only by the number of shades and the number of shuffling.

In addition, since the present invention is puzzled with an optimal number of shuffling times to obtain an appropriate puzzled image, the efficiency of operation can be maximized.

1 is a block diagram of a document security system for a document security method according to an embodiment of the present invention;
2 is a block diagram showing a modification of the document security system;
3 is a flowchart of a document security method according to an embodiment of the present invention.
4 is a flowchart illustrating a process of changing the image by the puzzling step S110 and the distributed storing step S130 and a process of changing the image by the combining step S220 and the reverse puzzling step S230 ).
FIG. 5 is an image showing the process of image variation as the number of shuffling increases in the padding step 110; FIG.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In the following description of the embodiments of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Referring to FIG. 1, a document security system for a document security method according to an exemplary embodiment of the present invention includes a plurality of DB servers 3 for dividing and storing documents in a distributed manner, document information for retrieving and collecting distributed stored documents And a document security server (6) for storing a document requested to be registered in the DB server (3) in a plurality of DB servers (3) ).

The document security server 6 divides a document into a plurality of DB servers 3 and stores the documents in a distributed manner when the document security DB 6 requests a document DB from the first terminal 1 through the network 5, A document registration unit 10 for registering and storing document information in the document information DB server 4 and a plurality of DB servers 3 And a document restoring unit (20) for transferring and retrieving the distributed stored document from each DB server (3) according to the document information stored in the document information DB server (4) to the second terminal (2) .

Here, the document to be distributed and stored in the plurality of DB servers 3 may be a document including only a text element or an image including only graphic elements, a graphic document, and of course, when the document is created as graphic data, Do not use it.

Documents can be classified into paper documents (including booklets) and electronic documents.

The electronic document includes a word document, a database file, a spreadsheet, a program source code, an image file (graphic data), and the like.

That is, according to the present invention, in the first terminal 1, an image composed of graphic data, which is a type of an electronic document, an image obtained by scanning a paper document, or an image obtained by format- To the document security server 6 or to convert the electronic document to an image in the document registration unit 10 of the document security server 6 when the document is transmitted from the first terminal 1 to an electronic document that is not an image file .

Therefore, the document registration unit 10 is preferably provided with a document-image conversion unit (not shown).

The document registration unit 10 for receiving such an image registration request includes a puzzling unit 11 for dividing an image into a plurality of pieces and then changing the arrangement of image pieces to obtain a puzzled image, And the image pieces of the puzzle image are distributed and stored in a plurality of DB servers 3, and the puzzle information, distributed storage information, and attribute information are stored as document information, And a DB formatting unit 13 for registering and storing the data in the DB server 4.

The attribute information of the document may include information for commonly used document classification and inquiry, meta information, and the like. In the embodiment of the present invention, in order to search the document contents by words or keywords, , Text included in OCR (Optical Character Recognition) is included in the attribute information, and the included text may be indexed, for example. Of course, the attribute information may be transmitted from the first terminal 1.

According to the embodiment of the present invention, after dividing positions of images are numbered, two numbers are generated by a pseudo random number generator in which a random number sequence is determined according to a seed value. The shuffling process of exchanging the image fragments of the numbered division positions is repeated a predetermined number of times so that the puzzle information includes the seed value and the shuffling count, But may not include all of them.

Meanwhile, as described later, according to the embodiment of the present invention, the shuffling count is determined by a convergence rate to a stochastic normal distribution and a function of the number of divided image fragments.

The distributed storage information includes an IP address or a storage location of the DB server 3 stored for each classified group, and is used as access information for loading through the network.

Information on the position (i.e., division position) on the puzzled image for the image pieces in each group may be included in the distributed storage information.

However, according to the embodiment of the present invention, the image distributed and stored in the plurality of DB servers 3 can be obtained by leaving the image fragments on the puzzled image for each group unchanged, leaving the remaining region empty, Filled images of scarcity. Therefore, the position information of the image pieces in each group may not be included in the distributed storage information. However, when the arbitrary image is filled in the sparse image, it is preferable that information for identifying an arbitrary image is included in the distributed storage information.

The document restoration unit 20 is a component that allows the second terminal 2 to retrieve or query the plurality of scarce images and document information stored as described above to receive a document, A retrieval unit 21 for obtaining puzzification information and distributed storage information corresponding to the requested document by specifying attribution information related to the requested document based on the attribution information when access is made via the network 5 and requesting document provision, A method for restoring an image to obtain a pre-puzzled image by rearranging a scarcity image distributed in the DB server (3) according to distributed storage information, rearranging the rearranged sequence in a puzzle based on the puzzled information after obtaining a combined image (22), and an electronic document obtained by pre-graphicization by OCR (Optical Character Recognition) when the obtained image is an image obtained by converting the document into a graphic form And a section (23).

According to the embodiment of the present invention, since the image distributed to the plurality of DB servers 3 is a scarcity image that does not change with respect to the position of the image fragment on the puzzled image, each scarcity image taken by each DB server 3 is superimposed You can restore the puzzle image.

FIG. 2 is a block diagram showing a modification of the document security system shown in FIG. 1, in which the document registration unit 10 is provided in the first terminal 1, the document restoration unit 20 is provided in the second terminal 2, And the document security server 6 is not provided separately. Here, the document restoring unit 20 can obtain a document from the scarcity image distributed and stored in the plurality of DB servers 3 according to the document information registered in the document information DB server 4. [

The document security server 6 includes the paging unit 11 and the attribute extracting unit 12 of the document registering unit 10 and the DB generating unit 13 in the first terminal, And a document restoration unit 20 may be provided.

Such a modified embodiment is merely a replacement of the constituent elements of the same function.

Hereinafter, a document security method in the document registration unit 10 and the document restoration unit 20 of the document security system will be described in detail.

3, a document security method according to an exemplary embodiment of the present invention includes a document registration step S100 performed by the document registration unit 10, a document registration step S100 when a document is requested to be provided from the second terminal 2 (S201) a document restoration step performed by the document restoring unit 20 (S200).

The document registration step (S100) includes dividing an image formed of an image of a document containing a text element or a graphic document into a plurality of fragments, and then repositioning the image fragments to change the arrangement thereof. (S110) for obtaining a puzzled image which prevents the shape of the document from being inferred from the image of the document or the document, A distributed storage step S130 of dividing the image pieces of the puzzled image into a plurality of groups and distributing the divided image pieces to a plurality of DB servers 3, And a document information DB making step (S140) of registering the information and attribute information in the document information DB server 4 as document information.

The puzzling step (S110) according to the embodiment of the present invention will be described with reference to FIG. 4 and FIG. Here, referring to FIG. 4, the change in the arrow direction of the solid line may be referred to.

As shown in FIG. 4, the puzzling step S110 is a process of dividing the original image A into a lattice shape, and then it is difficult to infer the original image A, Changing the arrangement of the divided images (a) until the image B is obtained, wherein according to the embodiment of the present invention, a shuffling process of randomly taking the two divided images (a) Repeat.

Referring to the drawings specifically illustrated in FIG. 5, the original image (A) is divided into seven divided images of seven columns and nine columns, and divided into 63 divided images. After that, the shuffling process of exchanging the image pieces (a) at positions corresponding to the two random numbers out of the numbers 1 to 63 is repeated by a predetermined number of shuffling times. As shown in FIG. 5, when the random numbers are different from each other in the three shuffling processes, the six divided images are moved and the divided images (A, B, the number of a) also increases.

However, the data amount of the puzzle information increases by the number of random numbers used to obtain the puzzled image. If the number of shuffling times is increased, the time required for the puzzling step (S110) becomes longer. If the number of shuffling times is too small There is a problem that an appropriate puzzled image can not be obtained.

In the embodiment of the present invention, the seed value and the shuffling count are used as puzzle information irrespective of the number of random numbers, a certain amount of data is provided without increasing the data amount due to the variation of the shuffling count, Also use the appropriate shuffling number to get an appropriate puzzled image.

First, for a random number, a random number sequence is generated by a pseudo random number generator whose random number sequence is determined according to a seed. And repeats the shuffling process of exchanging positions of image fragments of the divided positions. Of course, it is preferable to set the seed value to an arbitrary value.

The puzzle information includes the seed value and the appropriate shuffling count. That is, in the following reverse puzzling step S230, a seed value is applied to the same pseudo-random number generator to generate a random number sequence of two times the proper shuffling number, thereby obtaining a random number applied to the puzzle processing.

Next, with respect to the number of shuffling times, the convergence to a stationary distribution in which each image piece or one specific image piece can exist stochastically with a uniform probability irrespective of the dividing position the convergence rate and the number of divided image fragments.

Preferably, the degree of convergence is a normalized value obtained by dividing the maximum deviation of the probability of each divided position by the divided position probability at the normal distribution, so that the deviation of the probability distribution by position differs according to the number of divided image fragments Respectively.

Here, the split position probability at the normal distribution is the reciprocal of the number of image fragments.

The puzzling step (S110) calculates an appropriate shuffling count by the following equation (1).

In Equation (1), k is an appropriate shuffling number, N is the number of divided image pieces, and? Is a predetermined convergence degree as a normalized value as described above.

The derivation of Equation (1) will be described in detail below.

The attribute information extracted in the attribute extracting step S120 may include classification information for managing a document, inquiry information, meta information, and the like. In an embodiment of the present invention, An index, and the like are further extracted and included in the attribute information.

The distributed storage step S130 may be performed by leaving the image fragments a on the puzzled image B as unchanged for each classified group to generate scarcity images C1, C2, C3, and C4, and distributes the scarcity images C1, C2, C3, and C4 generated for each group to the different DB servers 3 in a distributed manner.

At this time, the scarcity images (C1, C2, C3, C4) for each group are left in a state in which the remaining region b is left empty except for the position where the image piece a is present, Fill with images. Here, any identifiable image is any image which can be discerned in order to delete the remaining region (b) except for the image fragment (a) in the collecting step (S220) described below.

For identification, for example, a method of adding a previously promised pattern to an arbitrary image can be used.

Of course, it is also possible to include the position of the image segment (a) in the scarcity images (C1, C2, C3, C4) for each group in the distributed storage information so as to be distinguishable from the remaining region (b).

Thus, each divisional image operation (a) on the puzzled image B exists and is not duplicated in any one of the plurality of scarity images C1, C2, C3, C4.

The distributed storage step (S130) generates distributed storage information including access information to the stored sparse images in the plurality of DB servers 3, that is, an IP address or a storage location (address) in the DB server 3 .

Meanwhile, it is possible to store the image pieces (a) as images with each other in each group, and to store the position information on the puzzled image (B) in each image piece (a) The complexity is low in restoring and restoring, so that the following description will be made on the assumption that the images are scattered and stored as a scarcity image.

The document information DB making step S140 is a step of converting the puzzle information (i.e., the seed value and the appropriate shuffling information) applied to obtain the puzzled image B from the image A of the original document in the puzzling step S110 The attribute information extracted in the attribute extracting step S120 and the distributed storage information which is the access information in the DB server 3 stored in each group in the distributed storage step S130 as document information, ).

After the document is distributed and stored as an image and the document information is registered, the document is requested to be viewed or downloaded from the second terminal 2 (S201) A restoration step (S200) is performed.

The document request at this time requests the image corresponding to the attribute information by transmitting the attribute information of the document, and the text containing the document contents can be used as the search keyword.

The document restoration step S200 includes a retrieving step S210 of obtaining corresponding puzzling information and distributed storage information from the document information DB server 4 on the basis of the attribute information of the document when requested to provide the document, A collection step S220 of collecting the stored scarcity images from the respective DB servers 3 according to the distributed storage information and obtaining a combined image, and rearranging them in the reverse order of the arrangement change when puzzling based on the puzzle information (S230) for obtaining an image of the document in the reverse puzzling step (S230), and a documenting step S240 for acquiring a text before the graphic recognition when the image obtained in the reverse puzzling step S230 is an image obtained by converting the document into a graphic form In order.

The searching step S210 can obtain the puzzle information and the distributed storing information by specifying the attribute information corresponding to the requested document according to the document classification information, the inquiry information or the meta information, but the text in the document is included in the attribute information , The attribute information can be specified by searching the text included in the attribute information as a search word at the time of the document request, and puzzification information and distributed storage information corresponding to the specific attribute information can be obtained.

The combining step (S220) and the reverse puzzling step (S230) are indicated by dashed arrows in Fig.

Referring to FIG. 4, the collecting step S220 creates a puzzled image B by superimposing the scarlet images C1, C2, C3 and C4 loaded from each DB server 3 according to the distributed storage information.

That is, since the image pieces (a) do not overlap in the different scarcity images (C1, C2, C3, C4), they can overlap each other to obtain the puzzled image (B). Of course, in the case of employing a method of filling the above-mentioned arbitrary image into the remaining area b except for the image fragment (a) in each of the scarity images C1, C2, C3 and C4, And then overlap.

The reverse puzzling step S230 is a step of obtaining the original image A before puzzling according to the seed value and the appropriate shuffling count included in the puzzle information and includes a pseudo random number generator generating two random number sequences of the shuffling number of times to obtain the same random number sequence as the puzzling step S110 and taking two random numbers in the reverse order of the random number sequence to generate image pieces of the two random number- Repeat shuffling process to exchange. Thus, the original image A before puzzling can be obtained.

The documenting step S240 extracts the text by performing optical character recognition (OCR) on the original image A obtained in the reverse puzzling step S230, and electronically documents the original image in the original document format. It is preferable that the format of the original document is put into the meta information of the attribute information or the scarcity image so that it can be known in the documenting step S240.

The documenting step S240 provides an electronic document such as a word document, a database file, a spreadsheet, a program source code or the like, which is made into an image DB, thereby widening the utilization range of the second terminal 2.

The electronic document or text thus obtained does not exist and the image that has not undergone the documenting step is transmitted to the second terminal 2. Alternatively, the image before the electronic documenting by the documenting step S240 may be transmitted.

Of course, when the first terminal 1 and the second terminal 2 are connected to the document security server 6, an authentication process is performed to secure connection security, and then a document registration and document restoration service should be provided.

Hereinafter, the basis of Equation (1) will be described.

First, the DTMC (Discrete Time Markov Chain) is defined as follows.

Definition 1: Discrete Time Markov Chain (DTMC)

If the Markov property is satisfied for the sequence X ₁ , X ₂ , ... of a random variable having a value in the countable state space S, this sequence is defined as DTMC.

<Markov property>

에 대하여, all

about,

이 n에 관계 없이 일정하면 time homogeneous DTMC 라고 한다. In the Markov property, the state of the (n + 1) -th viewpoint is determined by the state of the n-th viewpoint just before the instant,

If it is constant regardless of n, it is called time homogeneous DTMC.

를 i번째 row 및 j번째 column의 값을 갖는 N x N 크기의 transition matrix P=(p_ij)를 정의할 수 있다.If the size of state space S is N,

Can be defined as a transition matrix P = (p _ij ) of size N x N having the values of the i-th row and the j-th column.

Next, we introduce the theorem for Definitions 2 and 3 as follows.

<Definition 2: limiting distribution>

에 대하여 k번째 distribution 을

=

라고 정리할 수 있고, 이때

가 존재한다면 우리는 이 row vector를 limiting distribution이라고 정의한다.Let P be the transition matrix of the DTMC with state space S of size N. The initial distribution

The kth distribution for

=

In this case,

We define this row vector as the limiting distribution.

의 각 entry는 확률이기 때문에 0 이상 1 이하의 값을 갖고, entry의 합은 1의 값을 갖는다.In this definition, the initial distribution is the probability that a random variable X ₀ exists in each state. Therefore,

Each entry has a probability of 0 or more and 1 or less, and the sum of the entries has a value of 1.

<Definition 3: stationary distribution>

Let P be the transition matrix of the DTMC with state space S of size N,

를 stationary distrabution이라고 정의한다.A row vector of size N satisfying the condition

Is defined as stationary distrabution.

If a DTMC with a state space S of size N <∞ is irreducible and aperiodic, then there is a stationary distribution for the distribution π made up of N row vectors, and the limiting distribution is consistent with the stationary distribution regardless of the initial distribution. Can be obtained.

Therefore, we are interested in stationary distribution and can interpret distribution according to shuffling as follows.

As described with reference to Figs. 4 and 5, after the original image A is divided into N pieces and the sequence of the division pieces is sequentially ranked from 1 to N, A random number generator generating a uniform random number and generating a random number by repeating a shuffling process of mutually exchanging image fragments at two random numbers to generate a puzzle image B).

In this case, if the state of a particular image fragment is set to state, the state space S is {1,2, ..., N} since a particular image fragment can be moved to one of N positions each time shuffling is performed And the size of the transition matrix P can be thought of as N x N. That is, when the initial state is set to X ₀ = 1, the initial distribution is set to [1,0,0, ..., 0].

Thus, even if the position of one image segment selected by a specific image segment is set to a state, the distribution of other image segments according to an increase in the number of shuffling will also have a distribution substantially identical to that of a specific image segment. We will look at it in a limited way.

이다.If the position of a particular image fragment changes due to shuffling with two random numbers, since both of the two random numbers are either the current position of a particular image fragment or the current position of a particular image fragment,

to be.

가 된다.In addition, when the position of a specific image fragment is changed to another position by shuffling, since any one of the two random numbers is the current position of the specific image fragment and the other random number is the other position,

.

Thus, we can define the next transition matrix P with the entry P (i, j) as the entry to be changed from the current i position to the j position.

According to the transition matrix P described above, all entries are irreducible because they are positive numbers, and diagonal entries are both aperiodic because they are both positive. Also, the transition matrix P is a symmetric doubly stochastic matrix, since the sum of each row is always 1 and the sum of each column is always 1.

Thus, the transition matrix P is irreducible and aperiodic, so if we apply the foregoing, there is only one stationary distribution, and the limiting and stationary distributions are consistent regardless of the initial distribution.

이 stationary distribution임을 확인할 수 있다.And, since the sum of the columns of the transition matrix P is 1

This is a stationary distribution.

It can be considered that increasing the shuffling number gradually converges to a uniform distribution in which the probability that a specific image fragment at an initial specific position may be in any position of 1 to N is always the same regardless of the position.

Thus, the degree of convergence is considered to obtain the appropriate shuffling number.

In this case, the eigenvalue and eigenvector of the transition matrix P can be directly obtained as follows.

When eigenvalue? = 1,

Since the sum of each row is 0 at this time, the eigenvector is

And the normalized eigenvector

Can be obtained.

일 때에, Next, the eigenvalue

In this case,

. Looking at this matrix, all entries in the row are the same. That is, since row rank is 1 and nullity is N-1, linearly independent N-1 eigenvectors can be obtained.

에서 N개의 eigenvector를 모두 찾았으므로, diagonal decomposition을 이용할 수 있다. 또한, transition matrix P는 symmetric하므로 orthogonally diagonalizable하다. 따라서, Gram-Schmidt process를 통해 eigenvalue

에 대한 orthonormal한 eigenvector set e₂,e₃, ...e_N 를 구할 수 있고, 다음과 같이 matrix D와 Q를 정의하여 transition matrix P를 표현할 수 있다.As a result, two eigenvalues λ = 1 and

Since we have found all the eigenvectors in Eq. (1), we can use diagonal decomposition. Also, the transition matrix P is symmetric and orthogonally diagonalizable. Thus, through the Gram-Schmidt process, eigenvalue

We can obtain the orthonormal eigenvectors set e ₂ , e ₃ , ..., e _N for the matrix D and Q to define the transition matrix P as follows.

에 대해 k번째 transition(셔플링)한 상태의 distribution

을 다음과 같이 전개 가능하다.And, the initial distribution

The distribution with the k-th transition (shuffling)

Can be developed as follows.

Here, as described above, even if another initial state is used for the initial distribution, the calculation process and the result are the same.

는 linearly independent한 eigenvector e₁,e₂, ...e_N 를 이용하여 The initial distribution

Using a linearly independent eigenvector e ₁ , e ₂ , ... e _N

,

,

도 존재하게 된다.Can be developed,

.

This initial distribution can be assigned to the distribution after the k-th transition to develop as follows.

이 되므로,

이 되며, 이 row vector를 u 라고 정의하면, k번째 transition 한 상태의 distribution에 대한 식을 다음과 같이 된다.Here, if you directly calculate a ₁

Therefore,

If the row vector is defined as u, the equation for the distribution of the kth transition state is as follows.

로서 1보다 작으므로, k를 무한대로 보냈을 때에, 우변이 zero vector로 수렴하고, 결국,

도 u로 수렴함을 알 수 있다.Looking at this equation,

Is less than 1, so that when k is sent to infinity, the right side converges to zero vector,

Is converged to u.

The convergence rate (α) is defined as follows.

The norm in this case can be thought of as the L1 norm, which means the distance between distribution π and uniform distribution u.

에 대한 수렴도

는

Thus, the k-th shuffled distribution

Convergence to

The

Respectively.

를 만족하는 k, 즉, 셔플링 회수 k의 식으로 정리할 수 있다.Finally, any value or value of the appropriate degree of convergence

K, that is, the shuffling count k.

Therefore, the shuffling number k can be defined by the following formula (2) as the final result formula.

[0053] From the equation (2), it can be seen that the appropriate shuffling number can be determined according to the number of image fragments by empirically determining the appropriate convergence degree [alpha].

If we look closely at the equation (2) or the equation summarized by the convergence degree?, The degree of convergence? Is affected by the number N of image pieces. For example, even if k shuffling is performed the same, the larger the N, the greater the probability of the initial state is.

In other words, as described above, the use of L1 norm in calculating the convergence degree α decreases the probability value in the uniform distribution in each state as the value of N increases, but the convergence degree α is the same value regardless of The difference in distribution values between states becomes more pronounced when N has a relatively large value.

In order to solve this problem, the following normalized convergence degree β is defined.

Convergence β is defined as the ratio of the variance of the probability value in the current distribution to the probability 1 / N that all states have the same value when the distribution is uniform.

By defining the convergence degree β as described above, the smaller the value of β is set, the less the deviation of the probability value becomes, and the distribution becomes closer to the uniform distribution.

의 각 state에 대한 다음 관계식Then, the convergence degree beta at the shuffling number k is the kth distribution

The following relation for each state of

The following equation can be obtained.

From this equation, finally, Equation 3 for the shuffling number can be obtained.

Since the degree of convergence β indicates the degree of interlacing of the image fragments, that is, the degree of puzzling, if the appropriate value is set, the number of shuffling k for obtaining the convergence degree β according to the number of image fragments can be calculated Show.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, . &Lt; / RTI &gt; Accordingly, such modifications are deemed to be within the scope of the present invention, and the scope of the present invention should be determined by the following claims.

1, 2: terminal 3: DB server 4: document information DB server
5: Network 6: Document security server
10: Document registration part
11: puzzler 12: attribute extractor 13: DB editor
20:
21: searching unit 22: image restoring unit 23:

Claims (9)

A document registration unit 10 for dividing a document into a plurality of DB servers 3 and storing document information for collecting distributedly stored documents in the document information DB server 4, And a document restoring unit (20) for restoring the restored document information according to the document information,
A puzzling step (S110) of dividing an image created as a graphical document of a document into an image or a graphic document into a plurality of pieces and then changing the arrangement of image pieces to obtain a puzzled image;
An attribute extraction step (S120) of extracting attribute information of a document in a text form,
A distributed storage step (S130) of dividing the image pieces of the puzzled image into a plurality of groups and storing them in a distributed manner in a plurality of DB servers (3)
A document information DB step (S140) of registering puzzle information, distributed storage information and attribute information as document information in the document information DB server 4,
A document registration step S100 performed by the document registration unit 10; And
A retrieval step (S210) of obtaining corresponding puzzle information and distributed storage information based on the attribute information of the document when the document is requested to be provided;
A collecting step (S220) of collecting distributed images in accordance with the distributed storage information and obtaining a combined image,
A reverse puzzling step (S230) of rearranging the rearrangement of the rearrangement sequence to obtain an image before the puzzling,
A document restoring step (S200) performed by the document restoring unit (20);
&Lt; / RTI &gt; The method according to claim 1,
In the document restoring step S200, if the image obtained in the reverse puzzling step S230 is an image obtained by converting the document into a graphical form, the document restoring step S240 includes a document obtaining step S240 for obtaining a pre- . The method according to claim 1,
The attribute extracting step (S120) extracts the text in the document and includes it in the attribute information,
The retrieval step (S210) searches the text included in the attribute information to specify the attribute information, and then obtains the puzzle information and the distributed storage information corresponding to the specific attribute information. The method according to claim 1,
In the distributed storage step (S110), image fragments on the puzzled image are left unchanged for each group, a sparse image in which the remaining regions are left empty, and stored in different DB servers 3,
The collecting step (S220) superimposes the sparse images taken by each DB server (3). The method according to claim 1,
In the distributed storage step (S110), image fragments on the puzzled image are left unchanged for each group, a sparseness image in which the remaining region is filled with an arbitrary image is generated and stored in different DB servers (3)
In the collecting step S220, the scarcity image taken by each DB server 4 is deleted after the arbitrary image is deleted. The method according to claim 1,
In the puzzling step S110, a division number of the image is numbered, and a random number sequence of two times the shuffling number is generated by a pseudo random number generator in which a random number sequence is determined according to a seed A shuffling process of taking two images in order of a random number sequence and exchanging positions of image pieces of the two random numbered positions is repeated to generate a puzzled image,
Wherein the puzzle information comprises a seed value and a shuffling number,
In the reverse puzzling step S230, a random number sequence of two times the shuffling number is generated according to the seed value, and two pieces of the random number sequence are taken in reverse order of the random number sequence to exchange the image pieces of the two random numbers A document security method for obtaining an image prior to a puzzle by repeating a shuffling process. The method according to claim 6,
The shuffling count is a convergence rate to a stationary distribution in which one specific image fragment can be stochastically present at a uniform probability regardless of the dividing position as the shuffling process is repeated, A document security method determined by a function of the number of image fragments. 8. The method of claim 7,
Wherein the degree of convergence is a value obtained by dividing a maximum deviation of a probability by a dividing position by a dividing position probability at a normal distribution. The method according to claim 6,
The shuffling count k may be calculated according to the number N of image pieces and the degree of convergence?

Document security method.

Images