KR101620631B1 - Nuclear energy-related similar technical document search system and its method - Google Patents

Abstract

The method comprising the steps of: extracting a keyword from information about a target document; designating the target document as at least one classification system in a database using the keyword; determining a textual basis between the document stored in the at least one classification system and the target document Determining a degree of similarity, extracting an image included in the object document, designating the object document as at least one classification system of the database using the image, Based similarity between the target document and the target document, and retrieving a similar document using the text-based similarity and the image-based similarity.

Description

[0001] NUCLEAR ENERGY-RELATED SIMILAR TECHNICAL DOCUMENT SEARCH SYSTEM AND ITS METHOD [0002]

TECHNICAL FIELD [0001] The present invention relates to a system and a method for searching a similar technical document related to nuclear power, and more particularly, to a system and a method for searching similar technical documents necessary for examination among technical documents that have been reviewed using similarity of text and images will be.

Strategic items are goods, software and technologies that are available for the manufacture, development, use and storage of conventional and weapons of mass destruction and missiles as their means of delivery. These strategic items have certain restrictions on imports and exports, because they can harm international peace and national security.

Individuals or companies intending to import or export strategic goods must obtain export permission from the relevant licensing authority before importing or exporting. Even if they are not strategic goods, they may be exported if they have the risk of being transferred as strategic goods. Do.

In the case of Korea, the Minister of Commerce, Industry and Energy has specified the strategic materials in Table 2 and 3 of the Strategic Goods Import and Export Notification according to the principles of the Foreign Trade Law and the multilateral international export control system.

The strategic commodity import and export notice specifies technical characteristics such as the name of the item corresponding to the strategic item and the performance and specifications that the item having the item name should have in order to correspond to the strategic item.

However, the contents of the strategic goods specified in the strategic goods import and export notification are very extensive, and in particular, in the case of technology export cases, since the control standard is abstract and comprehensive, It is not easy to know whether it is a strategic item specified in the strategic goods import and export notification.

If there is any doubt as to whether or not a specific individual or corporation deals with a commodity or technology that the commodity or technology handled by the commodity is a strategic commodity, And the technical advisory group judges whether or not the applied material is a strategic commodity. The subjective judgment of the technical advisory team is inevitably involved, and the technical advisory group composed of the technical advisory group There is a problem that it is difficult to expect a consistent and accurate judgment result when the expert is changed.

In particular, when carrying out a project in which a large number of technical documents are transferred, such as nuclear power plant exports, all technical documents must be reviewed within a limited time, and in order to maintain the consistency or objectivity of the examination results, A search, comparison, and verification process is essential for the audit results.

However, when searching the past examination result in the existing database system, it is inevitable to perform a keyword search according to the title of the document or the name of the article specified by the applicant. This is because in the past examination data (for example, about 5,000) It can cause difficulties in retrieving documents to be compared or referenced.

Also, since there are many documents (drawings, etc.) composed only of pure images that do not include keywords at all in the examination subject technical documents, it may be difficult for the judge to search for the data to be referenced when performing only the search by the keyword.

Accordingly, in the present invention, information about text and images included in a document newly requested for examination is extracted, and a text similar to the new examination document is compared with information about the text extracted from the past technical document, And to provide a method and a system for searching the same.

The method comprising the steps of: extracting a keyword from information about an object document, designating the object document as at least one classification system in a database using the keyword, determining a textual basis between the document stored in the at least one classification system and the object document Determining a degree of similarity, extracting an image included in the object document, designating the object document as at least one classification system of the database using the image, Based similarity between the target document and the target document, and retrieving a similar document using the text-based similarity and the image-based similarity.

The step of extracting keywords from the information on the target document may include the steps of inputting information on the target document, converting the information into text,

And extracting the keyword from the text.

The information about the target document may include at least one of text and images.

The keyword may represent a word associated with nuclear power.

Wherein the step of designating the target document as at least one classification system of a database using the keyword comprises the steps of determining at least one classification system of the database by using a frequency of the keyword and a weight for at least one classification system of the database Determining at least one membership degree corresponding to the classification system, and designating at least one classification system of the target document using the at least one membership degree.

Wherein at least one of the classification systems of the database includes at least one of a reactor coolant, a reactor vessel, a safe injection and quench cooling system, a steam generating machine, a neutron flux monitoring system, a control rod assembly, a fuel assembly, chemical and volume control, It can include water, steam, revenues, gas, emergency diesel generators, reactor containment buildings, waste, surveillance, plant protection, drainage, power, fuel handling and transportation, and other systems.

Wherein the step of determining the text based similarity between the document stored in the at least one classification system and the target document comprises determining a text based degree of similarity between a document stored in the at least one classification system and a keyword frequency And calculating the degree of similarity.

Wherein the step of designating the target document as at least one classification system of the database using the image comprises the steps of: extracting a local feature point of the image; extracting a local feature point of the image using the feature point of the pre- Calculating a frequency of a matching point of an image stored in the database using local feature points of the image stored in the database and feature points of the images judged in the dictionary, Determining a similarity between the image and an image stored in the database using a frequency, determining a similar image from the image stored in the database using the similarity, Single Motorcycles may further comprise the step of acquiring the system.

The step of determining a similar image from the image stored in the database using the similarity may further include the step of filtering an image having a degree of similarity to an image not related to the stored examination separately.

Wherein determining the image-based similarity between the document stored in the at least one classification system and the target document comprises: determining a number of similar images between the document stored in the at least one classification system and the target document, And a step of calculating by using.

A database including at least one classification system and a document stored in the at least one classification system, a keyword extraction unit for extracting keywords from information on the target document, an image extraction unit for extracting an image from the information about the target document, A system management unit configured to designate the target document as at least one classification system of the database using the keyword and the image, a text-based similarity and an image-based similarity between the document stored in the at least one classification system and the target document, Based similarity degree and a similarity document using the image-based similarity degree, based on the text-based similarity and the image-based similarity.

The database may include matching point information configured by extracting and clustering images included in past examination documents.

The system designation unit may store at least one classification system of the target document specified using the keyword and the image in the database.

The similarity determination unit may store the text-based similarity and the image-based similarity in the database.

The document retrieval unit may store the similar document retrieved using the text based similarity and the image based similarity in the database.

Through the similar document retrieval system and method according to the present invention, the judge can quickly and consistently search, compare, and confirm documents to be compared or referenced.

1 is a flowchart illustrating a similar document retrieving method according to an exemplary embodiment of the present invention.
2 is a flowchart illustrating a text-based similarity determination method according to an exemplary embodiment of the present invention.
3 is a flowchart illustrating an image-based similarity determination method according to an exemplary embodiment of the present invention.
4 is a block diagram illustrating a similar document retrieval system according to an example of the present invention.
5 is an exemplary diagram for calculating cosine similarity between extracted images according to an example of the present invention.
FIG. 6 is an exemplary view of a database storing grid information of an extracted image according to an exemplary embodiment of the present invention. FIG.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a similar document retrieving system and method according to an exemplary embodiment of the present invention will be described in detail with reference to the accompanying drawings.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application .

The terms first, second and third, etc. may be used to describe various components, but such components are not limited by the terms. The terms are used for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second or third component, and similarly, the second or third component may be alternately named.

In this specification, "target document" refers to a document input for a new examination, and the target document may include strategic material or nuclear-related information in text or image form.

In this specification, "systematic" means a systematic classification based on a certain principle or rule.

As used herein, the term "similar document" refers to a document containing a keyword or an image similar to a target document, or an image similar to an embedded image, among the documents stored in the reviewed document or database.

Also, in this specification, the term "judge" means an individual or group of experts who substantially determine whether the target document is a strategic material or nuclear related document at the export licensing authority.

1 is a flowchart illustrating a similar document retrieval method according to an exemplary embodiment of the present invention.

Referring to FIG. 1, a similar document search method according to an exemplary embodiment of the present invention includes extracting a keyword from information on a newly input target document (S100), extracting a target document from at least (S120) determining a text-based similarity between a document stored in the database and a target document stored in the specified at least one sorting system (S120), extracting an image included in the target document S130), designating the target document as at least one sorting system in the database (S140) using the extracted image (S140), determining the image-based similarity between the document stored in the database and the target document in the specified at least one sorting system (S150). The similarity is determined using similarity between the text-based similarity and the image- And retrieving the document (S160).

2 is a flowchart illustrating a text-based similarity determination method according to an exemplary embodiment of the present invention.

Referring to FIG. 2, a step S100 of extracting a keyword from information on an object document according to an example of the present invention includes inputting information on the object document (S200), converting the input information into text Step S210 and extracting keywords from the converted text (S220).

In step S200 of inputting information on the target document, information on the target document may be input to a server or a personal computer through a scanner or a document reader. In addition, the input information can be stored in a database of a server or a personal computer. The target document may be in the form of an electronic document such as HWP, MS OFFICE, PDF or the like or a paper document. In general, the target document is a document input or requested for a new examination.

The information input in the step of converting the input information into text (S210) may be processed by OCR (Optical Character Reader) and converted into recognizable text. At this time, the converted texts can be stored in the form of a txt file in the database.

If the input document is an electronic document such as HWP, MS OFFICE, PDF, etc., text can be extracted directly from the target document without OCR processing. The extracted texts can be stored in a database in the form of a txt file. If the input document contains an image and the image also contains text, the text contained on the image can be extracted through OCR processing. For example, if the design document in the form of a JPEG file and the description of the design drawing are text in the target document in the form of a PDF file, the text included in the design drawing can be extracted by OCR processing of the design drawing. In general, when it is difficult to confirm whether the target document is large and the target document includes an image, the examiner can extract the text by performing OCR processing on all inputted target documents collectively.

In step S220 of extracting keywords from the converted text, a keyword related to nuclear power can be searched and extracted. Examples of keywords related to nuclear power include nuclear reactors, nuclear fuel, moderator, coolant, boron, cadmium, and control rods. Nuclear-related keywords are: (1) extract all words except the root and the ending from the translated text; And (2) can be extracted by excluding words that are not related to atomic power among the extracted words. In addition, the extracted keyword can be stored in the database together with the extraction frequency.

The step (S110) of designating the target document as at least one sorting system of the database using the keyword extracted from the target document according to an exemplary embodiment of the present invention may further include the following two steps: (1) Determining a degree of belonging to the classification system of the database using the frequency of the keyword and the weight for the classification system (S230); And (2) a classification system of the target document is designated using the determined degree of affiliation (S240).

The affiliation rate of the object document according to the classification system according to the example of the present invention can be calculated by the following Equation 1.

[Formula 1]

Specifically, the database can store weight information of three levels, for example, 1, 2, and 3, for each keyword included in each system. The larger the value of the weight, the higher the weight. The examiner can also directly assign the weight for each keyword included in each classification system.

In order to calculate the membership level of each classification system, first, the weight of the extracted keyword is searched by matching with the keyword extracted from the target document and the keyword stored in the database. Secondly, the frequency of extracted keywords is multiplied by the weight of the extracted keywords for each extracted keyword, and the total sum thereof is obtained. That is, the product of the keyword frequency and the keyword weight is summed for all extracted keywords. In this case, a weight value such as TF-IDF (Term Frequency-Inverse Document Frequency) may be used in addition to the frequency of extracted keywords. Finally, the total sum obtained in the above step is divided by the total sum of the weights of the respective systems.

For example, if the frequency of the reactor, coolant, vessel, generator, and pump are configured as shown in Table 1 below and the keywords extracted from the target document A are configured as shown in Table 2 below, , And the degree of belonging to the first line can be calculated by the following equation (2).

nuclear pile 2 Coolant 2 Vessel 0 generator One Pump One

keyword Line 1 Line 2 Line 3 Line 4 Line 5 nuclear pile 3 2 0 0 0 Coolant 3 One 0 One 0 steam One One 0 3 0 Vessel 0 3 One 0 0 Hydrogen 0 0 3 0 2 generator 0 0 0 0 3 Pump 3 0 0 0 0

[Formula 2]

A high degree of classification by classification system means that the target document contains a lot of nuclear-related keywords for the classification system. The numerical range of systematic affiliation can be set between 0 and 1. However, the numerical range can be adjusted through normalization as needed. In order to designate the classification system of the target document by using the determined degree of affiliation by the system, the search system can display the classification system of the document to the judge in descending order of the affiliation of each system. The reviewer can select the classification system of the target document according to the knowledge and experience of the reviewer by referring to the displayed list of membership degrees. In another example of the present invention, the system can automatically display the top three classification systems from the belonging list to the judges.

At least one of the classification systems stored in the database includes at least one of the following systems: reactor coolant, reactor vessel, safety injection and quench cooling, steam generator, neutron flux monitoring, control rod assembly, fuel assembly, chemical and volume control, It can include water, steam, revenues, gas, emergency diesel generators, reactor containment buildings, waste, surveillance, plant protection, drainage, power, fuel handling and transportation, and other systems. In addition to the classification systems listed above, the database may also include different classification systems depending on the system.

The search system can significantly reduce the number and time of documents to be searched by specifying the classification system of the target document. For example, the classification system of the target document is designated as the system 1, the system 2, and the system 3 as the calculation result of the classification degree by the classification system. In the database, there are 100 documents for the system 1 among the total 1,600 documents, The search system can only retrieve documents for each of the systems 1, 2, and 3 without having to search all the documents in the database.

If at least one classification line of the target document is designated, then in step S250, the text-based similarity between the document stored in at least one classification system and the target document is determined. The cosine similarity can be used as the text-based similarity. The degree of cosine similarity may be calculated using the keyword and keyword frequency of the document stored in at least one classification system and the keyword and keyword frequency of the target document.

The cosine similarity means the similarity between the vectors measured using the cosine of the angles between the two vectors of the inner space. Cosine similarity can be applied to any number of dimensions and can often be used to measure similarity in multidimensional amniotic space. For example, in an information retrieval, each word constitutes a dimension, and a document has a vector value represented by the number of times each word appears. That is, the keyword frequency of the target document and the keyword frequency of the document stored in the classification system of the target document in the database can be expressed as a vector. The cosine similarity between two documents can be calculated as the product of the inner product of each vector and the length of each vector.

For example, if the frequency of keywords in the target document is equal to Table 3 and the frequency of keywords in the document stored in the classification system of the target document in the database is as shown in Table 4, the similarity between the two documents can be calculated as shown in Equation 3 below .

nuclear pile 2 Coolant 2 Steam generator 0 Hydrogen One Pump One

nuclear pile Three Coolant One Steam generator One Hydrogen Three Pump 0

 [Formula 3]

The range of similarity can be set between 0 and 1, but different ranges are possible depending on the system. In calculating the similarity, the TF-IDF weight of the document may be used in addition to the keyword frequency of the document. In displaying the similarity calculation result to the reviewer, the search system firstly calculates the similarity between the documents stored in the target document and the database in the classification system of the target document, and then displays the sorted documents in order from the document having high similarity as a result have. The reviewer can select a similar document based on existing knowledge and experience by referring to the sorted similarity list. In another example of the invention, the system may automatically select a similar document if the similarity is greater than or equal to a predetermined value without the intervention of the auditor.

3 is a flowchart illustrating an image-based similarity determination method according to an exemplary embodiment of the present invention.

The step of extracting an image included in the target document according to an exemplary embodiment of the present invention (S130) may include a program for selecting image objects in a target document and storing the selected image objects in a separate file or a program for outputting each page of the document as a JPEG file . The type of the extracted image may include a layout, a photograph, a drawing, a table, a graph, and the like.

Referring to FIG. 3, a step (S140) of designating an object document as at least one sorting system of a database using an image extracted from an object document according to an exemplary embodiment of the present invention includes extracting local feature points of the image S310), calculating the frequency of matching points of the image using the regional feature points and the feature points of the images reviewed in advance (S320), calculating local feature points of the images stored in the database and feature points Calculating a frequency of a matching point of the image stored in the database using the frequency of the matching point, determining the similarity between the extracted image and the image stored in the database (S330) Determining a similar image from the image stored in the database using the similarity (S340), and It may include a step (S350) that specifies at least one classification system of the database corresponding to group similar images.

In step S310, the local feature points may be extracted using algorithms such as Scale-Invariant Feature Transform (SIFT) and Speed Up Robust Features (SURF). In general, the SIFT algorithm extracts feature points using Difference of Gaussian (DOG) and computes a 128-dimensional feature vector with a value between 0 and 255. In addition, feature points can be extracted using other detectors in addition to DOG. The number of feature vectors extracted by the DOG method is 256 ¹²⁸ , so each feature vector is hard to match completely.

(S320) of calculating a frequency of matching points of the image using the local feature points and the feature points of the images judged in advance, using the local feature points of the images stored in the database and the feature points of the images reviewed in the pre- The step of calculating the frequency of matching points of images stored in the database (S325) may be implemented using a Visual Word dictionary. That is, the Visual Word dictionary is a dictionary constructed by extracting and clustering minutiae extracted from the images included in past review documents in the database, comparing the minutiae points of the Visual Word dictionary with the minutiae points of images stored in the target image or database The frequency of the matching point can be calculated.

The Visual Word dictionary contains a 128-dimensional SIFT feature vector and can use the K-means algorithm for clustering. The number of clusters of the K-means algorithm can be specified in advance by the evaluator or the developer.

The step of determining the degree of similarity between the extracted image and the image stored in the database includes calculating the frequency of the matching point of the image extracted from the target document and the frequency of the matching point of the image stored in the database with the input information, . The cosine similarity represents the similarity between the vectors measured using the cosine of the angles between the two vectors of the inner space. Since the frequency of matching points of the image extracted from the target document and the frequency of matching points of the images stored in the database can be expressed as vectors, the cosine similarity between images is calculated by the inner product / (the product of the length of each vector) . The two vectors can be represented using TF-IDF weights.

5 is an example of calculating the degree of cosine similarity using the frequency of the matching point of the image extracted from the document to be judged A and the frequency of the matching point of the image extracted from the comparison object document B stored in the database. The SIFT is usually 128 dimensions, but in FIG. 5, the similarity is calculated by expressing it as a three-dimensional vector for convenience. In Fig. 5, the similarity is calculated assuming that one image is included in each of the A and B documents.

(S340) of retrieving a similar image from the image stored in the database using the determined similarity may include providing images to the examiner with the image having the similarity of the predetermined value or more, or automatically selecting the similar image as the similar image to the examiner . ≪ / RTI > In addition, the step S340 may include a step of filtering out images having a degree of similarity to the images that are not related to the stored examination separately. For example, if similar images are composed only of images that are not relevant to the examination, such as logos, background images, etc., the irrelevant images can be excluded from the examination by using the filtering process. Specifically, the minutiae points of the irrelevant images such as the logo background image are separately stored, and compared with the image to be judged, if the similarity is more than a certain value, it can be judged that it is not relevant to the examination.

The step of designating at least one classification system of the database corresponding to the searched similar image is to designate the system of the image included in the target document using the systematic information tagged to the image searched with the similar image according to the similarity Step < / RTI > The database may store information on images extracted from past review data as shown in FIG. 6 in order to provide systematic information tagged to the retrieved images.

The step of determining (S150) the image-based similarity between the document stored in the database and the target document in the at least one classification system according to an exemplary embodiment of the present invention is characterized in that the similarity between the document stored in the at least one classification system and the target document (S360) using the number of images and the total number of images of the target document. That is, if the classification system for the image of the target document is determined, the system can search for the document that contains the image most similar to the image included in the target document among the documents stored in the classification system specified in the database. The image-based similarity between the retrieved document and the target document can be calculated by the number of similar images / the total number of images of the target document.

The retrieval of the similar document using the text-based similarity and the image-based similarity according to an exemplary embodiment of the present invention may include displaying the result of the retrieval of the text-based similar document retrieval result and the image- can do. For example, the combined result can be set to (text-based similarity + image-based similarity) / 2.

Based on the results integrated into the next step, the judge can utilize the documents recommended as the most similar documents at the time of the new examination. In addition, the search system may re-execute the text-based similar document search and the image-based similar document search process. If a keyword that is not utilized in the systematic classification process is found during the document search process, it can be displayed to the reviewer.

The judge can select a keyword presented by the system, assign a new weight to each keyword, and store the new keyword and its weight in the database. Also, after the final screening is completed, the text, image, and grid information of the reviewed document can be stored in the database.

4 is a block diagram illustrating a similar document retrieval system according to an example of the present invention.

4, a similar document search system according to an exemplary embodiment of the present invention includes a database 460 including documents stored in at least one classification system and at least one classification system, A keyword extracting unit 410, an image extracting unit 420 extracting an image from information on the target document, a system for specifying the target document as at least one sorting system of the database using the keyword and the image, Based similarity determining unit 440 for determining a text-based similarity and an image-based similarity between the document stored in the at least one classification system and the target document, a similarity determining unit 440 for determining similarity based on the text- And a search unit 450 for searching a document.

The database 460 stores matching point information formed by extracting and clustering the images included in past examination documents in a database 460 in the form of a Visual Word dictionary, 440 may provide the matching point information when the matching point information is requested.

The system designation unit 430 designates at least one classification system of the target document by using the keyword extracted by the keyword extraction unit 410 and the image extracted by the image extraction unit 420, And stored in the database 460. The database 460 may provide the similarity level determination unit 440 or the document search unit 450 with a classification system of the stored target document in response to a request from the similarity level determination unit 440 or the document search unit 450 .

The similarity degree determining unit 440 may store the text based similarity and the image based similarity determined by the similarity degree determining unit 440 in the database 460 and the database 460 may include Based similarity degree and the image-based similarity degree to the document retrieval unit 450. [0064]

The document retrieval unit 460 may store the retrieved similar document in the database 460 using the text-based similarity and the image-based similarity. Although not shown in FIG. 4, the similar document retrieval system according to an exemplary embodiment of the present invention includes at least one classification system, a text based similarity, an image based similarity, a text based similarity, And a display unit for displaying a similar document retrieved using the base similarity to the reviewer.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, You will understand. The examples described above are therefore to be understood as being in all respects illustrative and not restrictive

400: target document 410: keyword extracting unit
420: image extracting unit 430:
440: degree of similarity determination unit 450:
460: Database

Claims (15)

Extracting a keyword from information on a target document;
Designating the target document as at least one classification system in the database using the keyword;
Determining a text based similarity between the document stored in the at least one classification system and the target document;
Extracting an image included in the target document;
Designating the target document as at least one classification system of the database using the image;
Determining an image-based similarity between the document stored in the at least one classification system and the target document; And
And searching for a similar document using the text-based similarity and the image-based similarity,
Wherein determining image-based similarity between the document stored in the at least one classification system and the target document comprises:
Calculating the image-based similarity using the number of similar images between the document stored in the at least one classification system and the target document and the total number of images of the target document. The method of claim 1, wherein extracting keywords from the information on the target document comprises:
Inputting information on the target document;
Converting the information into text; And
And extracting the keyword from the text. 3. The method of claim 2, wherein the information about the target document includes at least one of text and images. 3. The method according to claim 2, wherein the keyword indicates a word related to nuclear power. 2. The method of claim 1, wherein the designating the target document as at least one classification system in the database using the keyword comprises:
Determining at least one belonging degree belonging to at least one classification system of the database by using the frequency of the keyword and the weight for at least one classification system of the database; And
And designating at least one classification system of the target document using the at least one degree of belonging. 6. The system of claim 5, wherein at least one of the classification systems of the database includes at least one of a reactor coolant, a reactor vessel, a safety injection and quench cooling system, a steam generator, a neutron flux monitor, a control rod assembly, a fuel assembly, A similar document retrieval system including a cooling system, a cooling system, a turbine and a generator, a water supply, a steam, a plural, a gas, an emergency diesel generator, a reactor containment building, waste, monitoring, Way. The method of claim 1, wherein determining the text based similarity between the document stored in the at least one classification system and the target document comprises:
And calculating the cosine similarity using the keyword and keyword frequency of the document stored in the at least one classification system and the keyword and keyword frequency of the target document. 2. The method of claim 1, wherein the designating the target document as at least one classification system in the database using the image comprises:
Extracting local feature points of the image;
Calculating a frequency of matching points of the image using the local feature points and the feature points of the images reviewed in advance;
Calculating a frequency of matching points of images stored in the database using local feature points of the images stored in the database and feature points of the images reviewed in the dictionary;
Determining a degree of similarity between the image and an image stored in the database using the frequency of the matching point;
Determining a similar image from an image stored in the database using the similarity; And
And designating at least one classification system of the database corresponding to the similar image. 9. The method of claim 8, wherein determining similar images from images stored in the database using the similarity comprises:
And filtering the image having the degree of similarity equal to or greater than a predetermined value with the separately stored examination and the irrelevant image. delete A database comprising at least one classification system and a document stored in the at least one classification system;
A keyword extracting unit for extracting a keyword from information on a target document;
An image extracting unit for extracting an image from information on the target document;
A system specifying unit for specifying the target document as the at least one sorting system using the keyword and the image;
A degree of similarity determining unit that determines a text-based similarity and an image-based similarity between the document stored in the at least one classification system and the target document;
And a document retrieval unit retrieving a similar document using the text based similarity and the image based similarity,
Wherein the similarity determination unit calculates the image-based similarity using the number of similar images between the document stored in the at least one classification system and the target document and the total number of images of the target document. 12. The similar document retrieval system according to claim 11, wherein the database includes matching point information configured by extracting and clustering the images included in past examination documents. 12. The similar document retrieval system according to claim 11, wherein the system designation unit stores at least one classification system of the target document specified using the keyword and the image in the database. 12. The similar document retrieval system according to claim 11, wherein the similarity degree determination unit stores the text-based similarity and the image-based similarity in the database. The similar document retrieval system according to claim 11, wherein the document retrieval unit stores the similar document retrieved using the text-based similarity and the image-based similarity in the database.

Images