KR102516560B1 - Managing system for handwritten document - Google Patents

Abstract

The present invention relates to a system for managing handwritten documents. The system comprises: a preprocessing unit; a model generation unit; a character recognition unit; a keyword extraction unit; and a document management unit. According to the present invention, even handwritten characters in various handwritings can be recognized with excellent accuracy.

Description

Handwritten document management system {MANAGING SYSTEM FOR HANDWRITTEN DOCUMENT}

The present invention relates to a document management system, and relates to a management system for handwritten documents capable of more efficiently managing by recognizing and analyzing characters of handwritten documents and classifying documents into similar topic groups.

ECM (Enterprise Content Management), an enterprise-wide content management system, refers to a technology used to store, deliver, and manage content or documents related to processing tasks within an organization.

Recently, ECM is attracting attention in that the systematic management of it greatly contributes to strengthening corporate competitiveness and improving productivity in a situation where the amount and type of content inside a company is rapidly increasing.

As information technology progresses across all industries, companies are thinking about efficient management of their digital contents such as file systems, DM/XML, documents, media, and ERP (Enterprise Resource Planning).

On the other hand, individuals or companies store and manage receipts collected during economic activities as post-documentary documents for accounting processing such as expense reports and expenditure resolutions, or tax processing such as global income tax returns.

Due to the nature of the logistics/distribution industry, various receipts and supporting materials (receipt, receipt, etc.) generated in the distribution process such as logistics are exchanged in paper form.

Since the receipts stored and managed in this way are made of paper, there is a fundamental risk of damage to the original, there is a technical and economic burden to prevent exposure to loss and corruption, and the storage space for receipts must be increased in proportion to the increase in the amount of receipts stored. There is a problem with

In addition, the person in charge of an individual or company extracts and classifies information necessary for accounting or tax processing from a conventional receipt by handwriting input method, records it in a book, or inputs and stores it in a PC where an accounting management program is installed, making it inconvenient to extract information. There is a problem.

Accordingly, it is necessary to develop a technology capable of automatically recognizing and classifying handwritten documents.

Republic of Korea Patent Publication No. 10-2012-0092931 (2012.08.22) "ECM system and application data management method of the system" Korean Registered Patent No. 10-2149050 (2020.08.21.) "OCR-based document analysis system and method using artificial intelligence"

An object of a handwritten document management system according to an embodiment of the present invention is to provide a document management system capable of recognizing characters displayed on a document after imaging a handwritten document.

In addition, a handwritten document management system according to an embodiment of the present invention aims to provide a document management system that can recognize handwritten characters after generating and learning handwritten character images from language data. do.

In addition, the handwritten document management system according to an embodiment of the present invention is a document management system capable of more efficiently managing documents by selecting topics of document images in which handwritten characters are recognized and providing them to users. is intended to provide

The tasks of the present invention are not limited to the technical tasks mentioned above, and other technical tasks not mentioned will be clearly understood by those skilled in the art from the following description.

A handwritten document management system according to an embodiment of the present invention includes a pre-processing unit that receives text data from language data and stores learning text data generated by processing the text data in natural language; a model generating unit generating a handwritten character image by inputting a training image obtained by converting the training text data into an artificial neural network; a learning unit learning the handwritten character image; a character recognition unit that receives an input image and recognizes and outputs character information based on the learning data of the learning unit; a keyword extraction unit for extracting a keyword after calculating a ratio of a specific noun from the text information output from a plurality of input images; and a document management unit that selects a topic from the keyword extracted by the keyword extraction unit and stores the input image having the same topic in a folder having the same folder name as the topic, wherein the document management unit includes the input image Receives the change information of the folder name where is stored, and if the number of users using the same folder name as the proposed topic is greater than the number of users using a folder name that is not the same as the proposed topic, the topic proposed again The same folder name is proposed to users who use folder names that are not identical to the proposed topic, and the number of users who use folder names that are not identical to the proposed topic is If the number of users is less than the number of users, a keyword matching a folder name that is not the same as the proposed topic among extracted keywords is reselected as the topic, and then the folder name in which the input image is to be saved is proposed to the user again.
Here, the artificial neural network includes an encoder that receives the training image and extracts features by reducing dimensions, and a decoder that generates the handwritten character image different from the training image by increasing dimensions using the features and style vectors. It can be characterized in that it is an adversarial neural network including an auto-encoder and a discriminator.

delete

delete

Here, the keyword extraction unit may derive keywords by applying a TF-IDF algorithm.

Here, the pre-processing unit may extract the text data from the language data through XML parsing, classify the text data according to morphemes, normalize the text data, and classify and store only nouns through the natural language processing.

According to embodiments of the present invention, at least the following effects are obtained.

According to the handwritten document management system according to the present invention, after imaging a handwritten document, characters displayed on the document can be recognized.

In addition, by generating and learning handwritten character images from language data, handwritten characters in various handwritings can be recognized with excellent accuracy.

In addition, efficient management is possible by selecting the topic of handwritten documents and classifying, storing, and managing handwritten documents to which the same/similar topics are assigned.

In addition, it is possible to improve the topic by receiving feedback on whether users use the topic, so that documents can be managed more efficiently.

Effects according to the present invention are not limited by the contents exemplified above, and more various effects are included in the present specification.

1 is a conceptual diagram of a management system for handwritten documents according to an embodiment of the present invention.
2 is a block diagram of a management system for handwritten documents according to an embodiment of the present invention.
3 and 4 are explanatory diagrams of a pre-processing unit of a handwritten document management system according to an embodiment of the present invention.
5 is a block diagram of a model generation unit of a management system for handwritten documents according to an embodiment of the present invention.
6 is an explanatory diagram of a document management unit of a management system for handwritten documents according to an embodiment of the present invention.

Since the present invention can apply various transformations and have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. Effects and features of the present invention, and methods for achieving them will become clear with reference to the embodiments described later in detail together with the drawings. However, it should be understood that the present invention is not limited to the embodiments disclosed below, but may be implemented in various forms, and includes all modifications, equivalents, and substitutes included in the spirit and technical scope of the present invention.

Prior to the description, the terms described in the detailed description will be described. In the following embodiments, terms such as first and second are used for the purpose of distinguishing one component from another component without limiting meaning. Accordingly, it goes without saying that the first element mentioned below may also be the second element within the technical spirit of the present invention. Also, expressions in the singular number include plural expressions unless the context clearly dictates otherwise. In addition, terms such as 'include' or 'have' mean that features, numbers, steps, operations, components, parts, or combinations thereof described in the specification exist, and that one or more other features or components are present. It does not preclude the possibility of being added.

In addition, in the drawings, the size of components may be exaggerated or reduced for convenience of description. For example, since the size and thickness of each component shown in the drawings are arbitrarily shown for convenience of description, the present invention is not necessarily limited to the illustrated bar.

Hereinafter, an embodiment according to the present invention will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, the same or corresponding components are assigned the same reference numerals, and overlapping descriptions thereof will be omitted.

The handwritten document management system according to an embodiment of the present invention recognizes the characters of the handwritten document, analyzes them, and classifies the documents into similar topic groups, thereby enabling more efficient management of handwritten documents. of management system.

1 is a conceptual diagram of a handwritten document management system according to an embodiment of the present invention, and FIG. 2 is a configuration diagram of a handwritten document management system according to an embodiment of the present invention.

The handwritten document management system 1000 according to an embodiment of the present invention includes a pre-processing unit 100, a model generating unit 200, a learning unit 300, a character recognition unit 400, A keyword extraction unit 500 and a document management unit 600 are included.

The pre-processing unit 100 is a component that receives text data on a website and stores learning text data.

Specifically, the pre-processing unit 100 receives text data from one or more of a language dictionary, academic journal, and news scripter. In this embodiment, texts included in language dictionaries, journals, and news scripts are defined as language data.

In this embodiment, characters are recognized after converting a paper document into a digital format, where the paper document includes handwritten (handwritten) characters. Accordingly, when converting a paper document with handwritten characters into a digital format, public data is used to improve the accuracy of handwriting recognition, and thus text data is received from the language data.

In general, English has only 26 letters in total, but in Korean, it is not easy to collect data on the entire spelling because the number of syllables that can be combined is 11,172 when combined with initial, neutral, and final consonants. As a result, it takes a considerable period of time to collect and analyze all the texts that exist on the website, and it is not easy to collect and analyze structured/unstructured data. , receive text data from language data selected from one or more of the news scripters.

3 and 4 are explanatory diagrams of a pre-processing unit of a handwritten document management system according to an embodiment of the present invention.

The pre-processing unit 100 includes a text collection unit 110, a normalization unit 120, a tokenization unit 130, and a natural language processing unit 140.

The text collection unit 110 extracts text from language data and stores it in a database. The language data may be provided in an Extensible Mark-up Language (XML) format, and texts may be extracted by performing XML parsing on the language data in the XML format.

The normalization unit 120 removes phrases such as special characters (-, +, etc.), for example, information not related to search, among the extracted texts.

The tokenization unit 130 performs tokenization processing on the normalized texts in units of morphemes. At this time, the tokenization unit 130 excludes dependent nouns (eg, 'thing', 'number', 'time', etc.) composed of one word, even though they are morphemes.

The natural language processing unit 140 performs Part of Speech (POS) tagging by attaching a part of speech to each token using a natural language processing parser, which is a syntax analyzer that processes natural language for each morpheme. That is, after performing part-of-speech tagging on texts that have been normalized and tokenized in units of morphemes, only nouns are stored in the noun storage of the database, excluding texts tagged with other parts of speech other than nouns. This is referred to as learning text data in this embodiment. By using only nouns as learning text data, it is possible to reduce text data from becoming too voluminous, but it is also efficient in reading handwritten paper documents used in logistics/distribution structures.

Also, the image generator of the database generates a learning image obtained by converting the training text data into an image, and stores it in an image storage unit.

5 is a block diagram of a model generating unit of a management system for handwritten documents according to an embodiment of the present invention.

The model generation unit 200 is a component that generates a handwritten character image by inputting a training image of training text data into an artificial neural network.

The artificial neural network is a Generative Adversarial Network (GAN) and includes an auto encoder 210 and a discriminator 240 . The auto encoder 210 includes an encoder 220 and a decoder 230.

Specifically, the encoder 220 receives the training image and extracts the features of the image while reducing the dimensions. Then, the decoder 230 generates a handwritten character image different from the training image while increasing the dimensions again by using the features extracted by the encoder 220 and the style vector. The handwritten text image refers to a text image written by hand.

In other words, the encoder 220 extracts a feature while reducing the dimensions from the training image in which the digital character is displayed, and this feature means a feature of the digital character. Then, the decoder 230 reconstructs the image while increasing the dimension of the style vector according to each style by reflecting the characteristics of the extracted digital character. This image is a handwritten character image, which is a character image written by hand rather than a digital character.

In summary, in this embodiment, a handwritten character image is generated through the model generator 200 from a learning image obtained by imaging the learning text data generated from the language data in the preprocessing unit 100.

The learning unit 300 receives and learns handwritten character images. The learning unit 300 generates learning data by learning using a convolutional neural network (CNN).

The convolutional neural network can extract useful information for character prediction among the features of an image through a convolution operation, and can discriminate characters included in an input image based on the extracted features. Since the convolutional neural network is well known, a detailed description thereof will be omitted.

After receiving the input image, the character recognition unit 400 recognizes and outputs character and number information included in the input image based on the learning data of the learning unit 300 . Here, the input image includes handwritten characters, and means an image file obtained by photographing a simple paper receipt, a paper tax invoice, a receipt of receipt, a receipt confirmation, and the like.

The character recognition unit 400 recognizes a character through Optical Character Recognition (OCR). The learning unit 300 receives a handwritten character image and recognizes the character based on the learned learning data.

The input image includes both digital characters and handwritten characters, and in the case of digital characters, each image may have a different font or size, and even in the case of handwritten characters, the handwriting is different from each other. The character image is generated, and the character recognition unit 400 recognizes the character of the input image based on the learning data obtained by learning the handwritten character image in the learning unit 300, so the recognition rate is very high.

The keyword extraction unit 500 is a component that extracts keywords capable of selecting a subject of an input image from characters recognized by the character recognition unit 400 . In general, the amount of paper documents generated in the distribution/logistics process is vast, and most of the paper documents possessed by the same individual/company are similar in type. For example, it may be classified into a simple receipt, a tax invoice, a receipt, and the like.

When a plurality of input images are input to the character recognition unit 400 and characters are recognized, the keyword extraction unit 500 extracts main keywords from the recognized characters.

The keyword extraction unit 500 uses the TF-IDF algorithm for keyword extraction. The TF-IDF algorithm is an algorithm that numerically indicates how important a specific word is in a specific document when there are documents written by users. That is, the corresponding keywords are words occupying an important weight in the texts recognized by the text recognition unit. Since the TF-IDF algorithm itself is known, a detailed description thereof will be omitted.

6 is an explanatory diagram of a document management unit of a management system for handwritten documents according to an embodiment of the present invention.

The document management unit 600 selects and classifies topics from the keywords extracted by the keyword extraction unit 500, and stores input images classified into each topic for each topic.

At this time, when selecting a topic for the first time, the document management unit 600 selects a keyword having the highest ratio among keywords extracted by the keyword extraction unit 500 as a topic.

In addition, the document management unit 600 may create a folder (directory) having the same folder name (directory name) as the corresponding topic, store input images for each topic in the corresponding folder, and manage them.

Meanwhile, the document management unit 600 receives change information of the corresponding folder name. The document management unit 600 selects a topic and creates a folder having the same folder name to store input images. If the user determines that the folder name is not appropriate and modifies it to another folder name, the corresponding information is received. Furthermore, folder names of other users who suggested the same folder name, that is, the same topic, can also be received.

If the number of users using the same topic as the proposed topic is greater than the number of users using the modified folder name, the document management unit 600 proposes the corresponding topic as a folder name to the users using the modified folder name.

If the number of users using the proposed topic among the users is less than the number of users using the modified folder name, or there is no user, the document management unit 600 determines whether there is a keyword matching the changed folder name among the extracted keywords. Review. If there is a keyword that matches the changed folder name among the extracted keywords, the document management unit 600 re-selects the keyword as a topic and proposes it again as the folder name in which the input image will be saved. If there is no keyword that matches the changed folder name among the extracted keywords, the document management unit 600 selects the most common word among the changed folder names as a topic and proposes the same folder name again.

Through this, among input images possessed by multiple users, input images suggested for the same topic can be ultimately saved in the same folder name. Therefore, the efficiency of storing, managing, and searching for electronic documents within a company becomes very good.

As described above, according to the present invention, a handwritten document management system capable of more efficiently managing is provided by recognizing and analyzing characters of handwritten documents and classifying the documents into similar topic groups.

The use of all examples or exemplary terms (eg, etc.) in the present invention is simply to explain the present invention in detail, and the scope of the present invention is limited due to the examples or exemplary terms unless limited by the claims. it is not going to be In addition, those skilled in the art will know that various modifications, combinations and changes can be made according to design conditions and factors within the scope of the appended claims or equivalents thereof.

Therefore, the spirit of the present invention should not be limited to the above-described embodiments, and not only the claims to be described later, but also all ranges equivalent to or equivalent to these claims shall fall within the scope of the spirit of the present invention. would be said to belong.

1000: Management system for handwritten documents
100: pre-processing unit 110: text collection unit
120: normalization unit 130: tokenization unit
140: natural language processing unit 200: model generation unit
210: auto encoder 220: encoder
230: decoder 240: discriminator
300: learning unit 400: character recognition unit
500: keyword extraction unit 600: document management unit

Claims (5)

a pre-processing unit that receives text data from language data and stores learning text data generated by natural language processing of the text data;
a model generating unit generating a handwritten character image by inputting a training image obtained by converting the training text data into an artificial neural network;
a learning unit learning the handwritten character image;
a character recognition unit that receives an input image and recognizes and outputs character information based on the learning data of the learning unit;
a keyword extraction unit for extracting a keyword after calculating a ratio of a specific noun from the text information output from a plurality of input images; and
A document management unit that selects a topic from the keywords extracted by the keyword extraction unit and stores the input image having the same topic in a folder having the same folder name as the topic;
The document management department,
Receiving change information of the folder name in which the input image is stored;
If the number of users using the same folder name as the proposed topic is greater than the number of users using a folder name that is not the same as the proposed topic, again the folder name identical to the proposed topic is the same as the proposed topic. Suggest to users who use undefined folder names,
If the number of users using the same folder name as the proposed topic is smaller than the number of users using a folder name different from the proposed topic, matching with the folder name not identical to the proposed topic among the extracted keywords After re-selecting the keyword as the topic, suggesting it to the user as the folder name in which the input image will be saved
A management system for handwritten documents.
delete According to claim 1,
The artificial neural network,
An auto-encoder including an encoder receiving the training image and extracting features by reducing dimensions, and a decoder generating the handwritten character image different from the training image by increasing dimensions using the features and style vectors, and a discriminator. Characterized in that it is an adversarial neural network that
A management system for handwritten documents.
According to claim 3,
The keyword extraction unit derives keywords by applying the TF-IDF algorithm
A management system for handwritten documents.
According to claim 4,
The pre-processing unit,
Extracting the text data from the language data through XML parsing, normalizing the text data, classifying the text data by morpheme, and classifying and storing only nouns through the natural language processing
A management system for handwritten documents.

Images