KR20110039900A - Iamge data recognition and managing method for ancient documents using intelligent recognition library and management tool - Google Patents

Abstract

PURPOSE: An ancient documents management method using an intelligent recognition library is provided to effectively recognize text data of the ancient documents by constructing a knowledge-base managing module including image information related with the ancient documents. CONSTITUTION: A page structure corresponding to page layout information of an intelligent recognition library is generated by analyzing a page structure of an inputted ancient document image. The corresponding information is determined by calculating the correspondence score between an image block of the inputted ancient document and the page layout information using a block property which is offered by the intelligent recognition library. A suitable page structure is searched through the corresponded score.

Description

{Iamge Data Recognition and Managing Method for Ancient Documents using Intelligent Recognition Library and Management Tool}

The present invention provides a method for recognizing and processing old document image data using an intelligent recognition library and a management tool, comprising: a knowledge base management module including an image content DB, a character model DB, and a term DB having a plurality of image content information as a management tool; Intelligent recognition library consisting of an image preprocessing module and a structural analysis module for reproducing image content information using the information of the knowledge base management module and providing the reproduced image content information to a document image recognition and processing unit, preprocessing and initial analysis. A document image recognition and processing unit for performing a step, a physical layout analysis step, a document content structure analysis step, a document recognition step, the document image recognition and processing unit, a knowledge base management module, an intelligent recognition library, and a user The user screen (UI) to communicate It is also the field of intelligent recognition libraries and archives management tool, image data acquisition and processing methods utilizing characterized in that.

The document recognition system for the realization of the specialized search service of imaged old documents requires the combination of character recognition technology. The character recognition technology extracts features from input character data and classifies them into one character model through matching with several predetermined character models. It consists of a process. The character model, which is built in advance during the training process, is the same as the criterion that maps to a concept through recognition, so the recognition target data is limited or affects the performance of the character recognition depending on the characteristics of the character model.

In addition, the character recognition technology for the conventional document recognition system is difficult to fully algorithmize the information processing ability of the person, there are difficulties in terms of quality and quantity according to the difficulty of the character. In other words, the conventional technology is based on the intuition and experience of the excellent character recognition function of the human being, but the qualitative difficulties arise because the general method of objectifying, quantifying, and formulating the recognition process on the computer is not completed. do. For example, in the case of printed characters, difficulties arise due to inconsistencies in various font shapes, sizes, slants, etc., and difficulties in quantitative aspects of character recognition are due to the large amount of various characters. There is a problem that it is difficult to construct a system having a large memory capacity in engineering, and a problem that takes a long time to recognize a specific character on a system having a large memory capacity occurs. In particular, the problem is severe when the information content (information content) targets a large number of characters, for example, the number of characters (10 kinds) from English characters (52 species), Korean characters (11,172 species), furthermore, Chinese characters (About 50,000 kinds), the amount of information increases as it changes, which increases the size of the character to be remembered as the type of letters increases, and increases the size of the recognition device that compares and determines various possibilities. Cause.

The qualitative and quantitative difficulty of character recognition as described above is a factor that makes it difficult to implement general-purpose character recognition having a certain performance on all data. Therefore, in order to solve the problem of efficient text data of a large number of image data documents, it is necessary to mount additional modules that can be applied to existing document text systems in a form that minimizes the difficulty of character recognition. It is required to build a more accurate and efficient system by building a professional database corresponding to Moreover, when the image of the document to be processed has a long collection period, the document image quality is low and includes various document structures, and in case of large data generated by various fonts, the performance of character recognition is further reduced. Implementation requires more continuous R & D.

The present invention has been made to improve the problems according to the technology related to the method for recognizing and processing old document image data using the intelligent recognition library and management tools.

The conventional technology occurs because the general method of objectifying and quantifying the recognition process in the computer in terms of the quality of character recognition and formulating it in the form of an algorithm is not completed, and in the case of printed characters, various font shapes and sizes ( There is a problem in that it cannot provide a system with high quality character recognition ability caused by inconsistency of size, slant, etc. In terms of quantitative character recognition, it is possible to use a large amount of engineering capacity due to the large amount of various characters. The purpose of the present invention is to provide a solution to solve this problem because there is a problem that it takes a long time to recognize a specific character on a system having a large memory capacity and a system that has a large memory capacity.

The present invention is to realize the desired object as described above,

A knowledge base management module composed of an image content DB, a character model DB, and a term DB having a plurality of image content information as a management tool; An intelligent recognition library composed of an image preprocessing module and a structure analysis module for reproducing image content information using information of the knowledge base management module and providing the reproduced image content information to a document image recognition and processing unit; A document image recognition and processing unit for performing a preprocessing and initial analysis step, a physical layout analysis step, a document content structure analysis step, and a document recognition step; Recognizing and processing old document image data using an intelligent recognition library and a management tool, the document image recognition and processing unit, a knowledge base management module, and an intelligent recognition library and a user screen (UI) for communicating with a user. Give a way.

According to the present invention, a number of image content information related to an old document is a problem of manually inputting an index word corresponding to the content of each old document when constructing an index DB for providing information retrieval and image search service of an old document with image data. It solves the conventional problems by constructing a knowledge base management module having an intelligent recognition library and efficiently providing information management for text data of the imaged old documents inputted using the information provided by the knowledge base management module. It is effective.

The present invention relates to a method for recognizing and processing old document image data using an intelligent recognition library and a management tool, comprising: a knowledge base management module comprising an image content DB, a character model DB, and a term DB having a plurality of image content information as a management tool; ; An intelligent recognition library comprising an image preprocessing module and a structure analysis module for reproducing image content information using information of the knowledge base management module and providing the reproduced image content information to a document image recognition and processing unit; A document image recognition and processing unit for performing a preprocessing and initial analysis step, a physical layout analysis step, a document content structure analysis step, and a document recognition step; Recognizing and processing old document image data using an intelligent recognition library and a management tool, the document image recognition and processing unit, a knowledge base management module, and an intelligent recognition library and a user screen (UI) for communicating with a user. It is about a method.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings 1 to 13 showing an embodiment to which the present invention is applied.

First of all, as a management tool, the knowledge base management module having a plurality of image contents information is composed of a database (DB) having a plurality of data information related to the old documents such as the image contents DB, the character model DB, and the term DB. It provides a specialized knowledge base for data formation. Specifically, the image content DB of the knowledge base management module includes preprocessing information, initial analysis information, physical layout information, document content structure information, recognition information, and subtitle information. Configure to be able to create, modify, and delete. That is, the preprocessing information of the image contents DB includes the tilt correction information, the outline and the unnecessary area elimination information, the dot noise elimination information, the document outer noise elimination information, and the rotated image correction function information of the input image data documented document. The information for initial analysis has singular information and character model DB index information for recognizing primitive image data, and the page layout information has block position information, block alignment information, internal line alignment information, and inter-block position information. The document content structure information (Logical structure) has content label information, label order information, and block index information corresponding to the label. The recognition information has the term DB index information and the character model DB index information. The subtitle information has a title level. (Title level) information, font size information, font thickness information, level display style information, and character DB index information for title recognition. In addition, the character model DB has a character model list information, a character model information, a sample character image information, a character view information, the term DB is configured to have a term DB list information, term list information, term information, term search and additional information, In the construction of an index DB for providing information retrieval and image search service of old documents with image data, a knowledge base management module having a plurality of image content information related to old documents has to be input manually. Therefore, it is possible to convert the image data of the old document into text data of the old document to provide an effect that solves the conventional problems.

An intelligent recognition library comprising an image preprocessing module and a structural analysis module which reproduces image content information by using the information of the knowledge base management module and provides the reproduced image content information to a document image recognition and processing unit as follows. This is a compiled processing unit that contains code (function or class) to effectively convey the information required by the document image recognition and processing unit among the module information. It keeps the system stable by separating frequently used functions from the main function. It is easy to debug and has configuration and effects to make compilation time faster.

That is, the image content information reproduced by the image preprocessing and structure analysis module of the intelligent recognition library using the information of the knowledge base management module may be information for initial analysis, page structure information, document content structure information, image data recognition information, and subtitle information. In addition, the information may be effectively transferred to the document image recognition and processing unit in each database of the knowledge base management module by a separate main function. In addition, the image preprocessing module of the intelligent recognition library is compatible with and reproduces the information for initial analysis consisting of image tilt correction information, image outline and unnecessary area removal information, and dot noise removal information of the image content DB of the knowledge base management module. Provide information for preprocessing and initial analysis of image archives in the preprocessing and initial analysis steps of the document image recognition and processing unit; The structure analysis module of the intelligent recognition library includes page structure information, document content structure information, recognition information, and subtitle information of an image content DB of a knowledge base management module; Character model list information, character model information, sample character image information, and character view information of the character model DB; Page layout (Physical layout) analysis step, Logical structure analysis step of document image recognition and processing unit by compatible and reproducing term DB list information, term list information, term information, term search and additional information of term DB By providing the document recognition step, the frequently used functions are separated from the main functions, so that the system can be stably maintained, easier to debug, and faster to compile.

In addition, the document image recognition and processing unit that performs text data of the imaged old documents as the preprocessing and initial analysis stage, the physical layout analysis stage, the logical structure analysis stage, and the document recognition stage. The preprocessing and initial analysis stages consist of document image preprocessing, image structure analysis, and primitive image data recognition process. By checking the formal state and form of the documented old document, it removes the obstacles of the recognized data recognition. This is a step to analyze the structure of the data-formed ancient document and to recognize it first, to make more accurate text data-old document. That is, the document image preprocessing process is performed by using the preprocessing information and the initial analysis information of the image content information of the image preprocessing module of the intelligent recognition library to correct the skew of the image data of the old document, remove dot noise, and outline the outline of the image data of the old document. And eliminating unnecessary areas, removing noise from the outside, and automatically correcting old documents with 90-degree rotated image data. The image structure analysis process uses page structure information among the image content information of the structure analysis module of the intelligent recognition library. In order to increase the accuracy of extraction of blocks, which is a basic element, it automatically extracts the blocks in consideration of the form of the old document page, and provides the function to identify and extract the picture and table areas in order to prevent the blocks between pictures, tables and texts from being combined. The initial image data recognition process is processed and analyzed by the above processes. It is a process of preparing the processing of the following physical layout analysis step by first recognition of the documented old image data.

In addition, the page layout analysis step of the document image recognition and processing unit is a process of finding page information corresponding to the image data input page, and analyzes the page structure of the inputted image data old document and provides it in the structure analysis module of the intelligent recognition library. A page structure generation process of generating a corresponding page structure among pages provided by the physical layout information; A correspondence information determination process of determining correspondence information by calculating a correspondence score between the page layout (Physical layout) information using the attributes of the block provided by the structure analysis module of the intelligent recognition library and the blocks in the input imaged old document page; A page structure search process for finding an appropriate page structure through the corresponding scores between the blocks calculated above; A page search process of finding a page corresponding to the page among physical layout information provided by the structure analysis module of the intelligent recognition library; Blocks related to subtitles during the page structure generation process, the corresponding information determination process, the page structure search process, and the page search process are separately managed, and this step is the system of the present invention most suitable for the input image data structured page structure. In order to select only the characters of the imaged data by searching the page structure stored in the structure of the input block to be displayed as a character of the page of the imaged old document.

The page structure generation process of generating the page structure corresponding to the page structure in the page layout (Physical layout) information provided by the structure analysis module of the intelligent recognition library by analyzing the page structure of the inputted image data old document in association with the above Since the block structure of the old document is different from the page block structure of the physical layout information provided by the image preprocessing module of the intelligent recognition library, the block structure information generated in the preprocessing and initial analysis stages is not perfect. Although the block must be extracted as one block, it is divided into several blocks and the correct correspondence information cannot be found between the blocks. Therefore, it is executed to determine the most similar correspondence between the blocks of two pages. Two or more blocks are merged to find a page having the most similar form of block structure between the block structure of the imaged old document page and the page block structure of the physical layout information provided by the structure analysis module of the intelligent recognition library. Allow it to configure the block structure to produce any type of page structure that can be produced. In this case, if the blocks to be combined have too different block attributes, the corresponding points are subtracted by calculating a penalty when calculating the corresponding points between the blocks as follows.

In addition, the present invention, after performing the above process, by calculating the corresponding score between the page layout (Physical layout) information using the properties of the block provided by the structure analysis module of the intelligent recognition library and the block in the input imaged old document page corresponding information The process of determining the correspondence information is performed. This process uses the attributes of font thickness, font size, and positional relationship between blocks to give the best response score when the attributes of two blocks match exactly. As a method of _{penalizing the} score, the similarity (S _AiBj ) of the block is calculated by using a calculation method as follows.

[Equation 1]

The S _PropertyAiB represents the similarity of the two corresponding blocks using the font size and the font thickness, which are the basic attributes of the block, so that the property similarity between the i-th block of page A and the j-th block of page B is S, and S _NeighborAiBj corresponds to The similarity of neighboring blocks i and j is shown by indicating how similarly the neighbors of the four blocks in the up, down, left, and right directions are located. The following equation is used to obtain the property similarity diagram S _PropertyAiB and the neighbor similarity diagram S _NeighborAiBj .

[Equation 2]

(The above fs _Ai and fs _Bj are the representative font sizes of blocks, and ft _Ai and ft _Bj represent the representative font thickness of each block.)

&Quot; (3) "

(S _NorthAiBj , S _SouthAiBj , S _EastAiBj , S _WestAiBj represent similarity in four directions of up, down, left, and right around i and j blocks.

&Quot; (4) "

W _Ai and H _Ai in Equation 4 indicate the width and height of the i-th block of page A, respectively, and W _AiNeiN , W _AiNeis , H _AiNeiE , and H _AiNeiW respectively indicate the top and bottom of the block i. Shows the width and height of neighboring blocks in the left and right four directions. Similarly, W _Bj , H _Bj , W _BjNeiN , W _BjNeiS , H _BjNeiE , and H _BjNeiW represent information of the j th block of page B. OverlapH () and OverlapV () represent a function for calculating the length of a section where two blocks overlap each other horizontally and vertically, and Ps represents a weight (similarity of overlapping neighbor blocks, 1 or 0.8).

In addition, the page structure search process of finding the appropriate page structure through the corresponding scores between the blocks calculated above is a process of finding the most suitable page structure by summing each page structure score based on the corresponding scores between the blocks. At this time, comparing the page structure scores may have an effect of finding a more suitable page structure by varying the weight of the page structure scores according to the size of the area of each block.

The page search process for finding a page corresponding to the physical layout information provided by the structural analysis module of the intelligent recognition library best corresponds to the image data-formed document page among the image content DB pages in the system of the present invention. In the process of finding the image content DB page, the page matching score between the image data documented document page and the image content DB page input using the calculated page structure score is added to the similarity score of each block and The image content DB page given the highest page matching score is selected.

In addition, a subtitle of each page of the inputted image data document during the page layout (Physical layout) analysis step of the document image recognition and processing unit consisting of the page structure generation process, the corresponding information determination process, the page structure search process and the page search process Blocks and their contents are managed separately from other contents and blocks. That is, a subtitle possibility test is performed on all lines during the generation and determination process by using subtitle information in the image contents information of the intelligent recognition library, and when a title determined to be a subtitle is found, the page structure search is divided into separate blocks. By reflecting and managing the process and the page search process separately, it is possible to find a more suitable textized index term for a full-text search.

Logical structure analysis step of document image recognition and processing unit corresponds to each page in order to compose the full text data of the imaged old document input after performing the physical layout analysis step as shown in FIGS. As a step of assigning the order of the block extraction result, each logical label included in the block index information and the content label information corresponding to each label among the document structure information provided by the structural analysis module of the intelligent recognition library. Each block of each page of the imaged ancient document input using the corresponding block information and each block of the page of the page of physical layout information using the properties of the block provided by the structure analysis module of the intelligent recognition library (Section title), content block) and the structure of intelligent recognition library The order in which the actual block (block of the input imaged old document) corresponding to each label among the document content structure information is determined and read using the label order information among the document structure information provided by the module. By determining the order of the blocks, the document content structure analysis is performed, which assigns a logical order to each block of the inputted imaged document page.

The document recognition step of the document image recognition and processing unit performs textual data recognition on all blocks of the inputted imaged old document page given a logical order, and each block of the inputted imaged old document page is intelligently recognized. Among the information for recognizing the image contents DB of the knowledge base management module reproduced from the library, the terminology of the old document imaged using the information of the character model DB and the term DB of the knowledge base management module using the term DB index information and the character model DB index information. Recognize data as textualized data. Also, in the document recognition step, the section title is title level information, font size information, font thickness information, level display style information, title recognition character DB index among information of the image contents DB of the knowledge base management module. Using the information of the character model DB and terminology DB of the knowledge base management module using subtitle information such as information, the data of the imaged old document is recognized as textualized data, and the documented step is carried out in the textual old document. The recognition result can be corrected to improve the recognition result by using the information of the recognition. The recognition candidate and recognition score are given to each recognition word in the recognition result, and the term DB stores weights for each word, thereby recognizing candidates of the recognition result. By adjusting the recognition score of, it is possible to increase the frequency of occurrence of words stored in the term DB. In addition, the character model learner used in the character model DB is a minimum distance classifier method using a quadratic distance function (QDF) among the probability statistical methods, and the quadratic distance function (QDF) and the mahalanobis distance function. Can be used in combination.

In addition, the present invention further includes a user screen (UI) for communicating the knowledge base management module, the intelligent recognition library, and the document image recognition and processing unit and the communication between the user, thereby enabling smooth communication between the system and the user. That is, the user screen (UI) is a thumbnail screen of an image data-formed old document to be input as shown in FIG. 12, an image contents DB list screen for checking page structure information among image contents information, and a screen for checking information of the corresponding image contents DB. It is possible to effectively manage and set a large amount of input image data for the recognition and processing of the image data old document to be inputted by configuring the image data, and the thumbnail screen of the input image data old document as shown in FIG. By constructing a screen, etc., it is possible to confirm the textification of the imaged old document.

1 is a summary block diagram of a system according to an embodiment of the present invention.

2 is a detailed block diagram of a system in accordance with an embodiment of the present invention.

Figure 3 is a block diagram of a module for managing the image content DB of the knowledge base management module according to an embodiment of the present invention.

Figure 4 is a block diagram of a module for management of the character model DB of the knowledge base management module according to an embodiment of the present invention.

Figure 5 is a block diagram of a module for the management of the term DB of the knowledge base management module according to an embodiment of the present invention.

Figure 6 is a block diagram showing a page and the page configuration selected in the page configuration generation process of the page layout (Physical layout) analysis step according to an embodiment of the present invention.

FIG. 7 is a block diagram illustrating a corresponding score calculated for each block in a process of determining corresponding information in a physical layout analysis step according to an embodiment of the present invention; FIG.

8 is a block diagram showing a process of finding a page configuration that is appropriately matched with a page configuration search process in a page layout analysis step according to an embodiment of the present invention.

FIG. 9 is a diagram illustrating a process of finding a page correspondingly to a page search process in a physical layout analysis step according to an embodiment of the present invention. FIG.

FIG. 10 is a block diagram illustrating a process of extracting a logical order of blocks in a logical structure analysis step according to an embodiment of the present invention. FIG.

FIG. 11 is a block diagram illustrating a process of extracting a logical order of blocks in a logical content analysis step according to an embodiment of the present invention. FIG.

Fig. 12 is a block diagram showing a recognition information setting user screen for document image recognition and processing among user screens UI according to an embodiment of the present invention.

Fig. 13 is a block diagram showing a document image recognition and processing result screen in a user screen UI according to an embodiment of the present invention.

Claims (10)

A knowledge base management module composed of an image content DB, a character model DB, and a term DB having a plurality of image content information as a management tool; An intelligent recognition library comprising an image preprocessing module and a structure analysis module for reproducing image content information using information of the knowledge base management module and providing the reproduced image content information to a document image recognition and processing unit; A document image recognition and processing unit for performing a preprocessing and initial analysis step, a physical layout analysis step, a document content structure analysis step, and a document recognition step; Recognizing and processing old document image data using an intelligent recognition library and a management tool, the document image recognition and processing unit, a knowledge base management module, and an intelligent recognition library and a user screen (UI) for communicating with a user. Way. The method of claim 1, The image content DB of the knowledge base management module includes information for preprocessing and initial analysis, physical layout information, document content structure information, recognition information, and subtitle information; And all the information is generated, modified and deleted. The method of recognizing and processing old document image data using an intelligent recognition library and management tool. The method of claim 2, Pre-processing information and initial analysis information include document skew correction information, outline and elimination of unnecessary lines, dot noise elimination information, document outer noise elimination information, rotated image correction function information, singular information and initial image data character model DB index information is included; Page layout information (Physical layout) includes block position information, block alignment information, alignment information of internal lines, and positional relationship information between blocks; The document content structure information (Logical structure) includes content label information, label order information, and block index information corresponding to the label; The recognition information includes the term DB index information and character model DB index information; The subtitle information includes title level information, font size information, font thickness information, level display style information, and character DB index information for title recognition. Recognition and processing method. The method of claim 1, The character model DB of the knowledge base management module includes character model list information, character model information, sample character image information, and character view information; And all the information is generated, modified and deleted. The method of recognizing and processing old document image data using an intelligent recognition library and management tool. The method of claim 1, The term DB of the knowledge base management module includes term DB list information, term list information, term information, term search and additional information; And all the information is generated, modified and deleted. The method of recognizing and processing old document image data using an intelligent recognition library and management tool. The method of claim 1, The image content information of the intelligent recognition library includes information for initial analysis, page structure information, document content structure information, image data recognition information, and subtitle information, and generates image content information using data of a knowledge base management module. A method for recognizing and processing old document image data using intelligent recognition libraries and management tools. The method of claim 1, The image preprocessing module of the intelligent recognition library regenerates the initial analysis information consisting of image tilt correction information, image outline and unnecessary area elimination information, and dot noise elimination information from the knowledge base management module. Method for recognizing and processing old document image data using an intelligent recognition library and management tool, which is provided to the preprocessing and initial analysis stage of the processing unit. The method of claim 1, The structure analysis module of the intelligent recognition library includes: page structure information, document content structure information, recognition information, and subtitle information of the image content DB of the knowledge base management module; Document image recognition and processing unit by regenerating the character model list information, character model information, sample character image information, character view information, term DB list information, term list information, term information, term search and additional information of the character model DB. A method for recognizing and processing old document image data using an intelligent recognition library and management tool, which is provided in a physical layout analysis step, a logical structure analysis step, and a document recognition step. The method of claim 1, The preprocessing and initial analysis steps of the document image recognition and processing unit include the document image preprocessing process, the image structure analysis process, and the initial image data recognition process. . The method of claim 1, The physical layout analysis step of the document image recognition and processing unit is a process of finding page information corresponding to the image data input page. A page structure generation process of analyzing a page structure of an input image data-old document and generating a page structure corresponding to the page structure in the page layout information provided by the structure analysis module of the intelligent recognition library; A correspondence information determination process of determining correspondence information by calculating a correspondence score between page layout (Physical layout) information using a block attribute provided by a structure analysis module of the intelligent recognition library and a block in an input imaged old document page; A page structure search process for finding an appropriate page structure through the corresponding scores between the blocks calculated above; A page search process of finding a page corresponding to the page among physical layout information provided by the structure analysis module of the intelligent recognition library; The method for recognizing and processing old document image data using an intelligent recognition library and a management tool, wherein a block relating to a subtitle is separately managed during the page structure generation process, corresponding information determination process, page structure search process, and page search process. .

Images