KR102619368B1 - Apparatus, method and user device for recognizing optimal text using ocr ai algorithm of a different kind based on deep learning - Google Patents

Abstract

An apparatus for recognizing text from an image includes a receiving unit that receives image data including text from a user terminal, and a text recognition engine selection unit that selects at least two text recognition engines including algorithms to be applied to the image data from the user terminal. , a candidate text derivation unit that derives each candidate text recognized from the image data using the selected at least two text recognition engines, and inputs each derived candidate text into a pre-trained reinforcement learning model to produce a final text. It includes a final text deriving unit that derives the final text and a providing unit that provides the derived final text to the user terminal.

Description

Device, method, and user terminal for recognizing optimal text using deep learning-based heterogeneous OCR AI algorithm {APPARATUS, METHOD AND USER DEVICE FOR RECOGNIZING OPTIMAL TEXT USING OCR AI ALGORITHM OF A DIFFERENT KIND BASED ON DEEP LEARNING}

The present invention relates to an apparatus, method, and user terminal for recognizing text from an image using a heterogeneous text recognition engine including an OCR AI algorithm.

OCR (Optical Character Reader) refers to a device that reads characters created in a document using light. OCR allows characters to be read through comparison between recognized characters and previously stored text data, but has the disadvantage of having some errors in recognizing the generated characters due to the reading characteristics of OCR.

In relation to this OCR function, Korean Patent Publication No. 2008-0002084, a prior art, discloses a system and method for optical character reading.

OCR uses special reading devices, but it can also recognize letters using small optical scanners and dedicated software. Previously, letters could be recognized through PC solutions that included logic, patterns, and algorithms, but recently, with the development of AI (Artificial Intelligence) technology, image learning has been implemented using machine learning and deep learning techniques to recognize scanned characters. By analyzing images, it became possible to automatically recognize letters.

However, various algorithms such as DBN, CNN, RNN, and GAN are used as neural networks for OCR AI. Because of this, as different text recognition results are derived depending on the algorithm used in OCR AI, it has the disadvantage that the user's satisfaction with text recognition varies depending on the algorithm used by OCR AI.

An object is to provide an apparatus, method, and user terminal that receive image data including text from a user terminal and select at least two text recognition engines including algorithms to be applied to the image data.

Each candidate text recognized from the image data is derived using at least two text recognition engines, each derived candidate text is input into a pre-trained reinforcement learning model to derive the final text, and the derived final text is sent to the user. We intend to provide a device, method, and user terminal provided as a terminal.

However, the technical challenges that this embodiment aims to achieve are not limited to the technical challenges described above, and other technical challenges may exist.

As a means for achieving the above-described technical problem, an embodiment of the present invention provides at least two or more devices including a receiving unit for receiving image data including text from a user terminal, and an algorithm to be applied to the image data from the user terminal. A text recognition engine selection unit that selects a text recognition engine, a candidate text derivation unit that derives each candidate text recognized from the image data using the selected at least two text recognition engines, and a candidate text derivation unit that derives each candidate text recognized from the image data using the selected text recognition engine. A text recognition device can be provided, including a final text deriving unit that inputs a learned reinforcement learning model to derive the final text, and a providing unit that provides the derived final text to the user terminal.

Another embodiment of the present invention includes selecting an image to recognize text from among a plurality of images, transmitting the selected image to a text recognition device, and at least two text recognition engines including an algorithm to be applied to the selected image. and receiving a final text derived from the at least two or more selected text recognition engines from the text recognition device, wherein the final text uses the at least two or more algorithms selected by the text recognition device. Thus, it is possible to provide a text providing method in which each candidate text recognized from the image is derived, and each derived candidate text is input into a pre-trained reinforcement learning model and derived.

Another embodiment of the present invention includes at least two images including an image selection unit that selects an image to recognize text from among a plurality of images, a transmission unit that transmits the selected image to a text recognition device, and an algorithm to be applied to the selected image. a text recognition engine selection unit that selects one or more text recognition engines; and a provision unit that receives a final text derived from the text recognition device through the at least two selected text recognition engines, wherein the final text is provided by the text recognition device. Provided is a user terminal in which each candidate text recognized from the image is derived using the selected at least two text recognition engines, and each of the derived candidate texts is input into a pre-trained reinforcement learning model and derived. You can.

Another embodiment of the present invention includes receiving image data including text from a user terminal, selecting at least two text recognition engines including an algorithm to be applied to the image data from the user terminal, and selecting the selected text recognition engine. Deriving each candidate text recognized from the image data using at least two text recognition engines, inputting each derived candidate text into a pre-trained reinforcement learning model to derive a final text, and deriving the final text. A text recognition method may be provided including providing the final text to the user terminal.

The above-described means for solving the problem are merely illustrative and should not be construed as limiting the present invention. In addition to the exemplary embodiments described above, there may be additional embodiments described in the drawings and detailed description of the invention.

According to one of the means for solving the problem of the present invention described above, in the past, different text recognition results were derived depending on the algorithm used for learning in OCR AI, but by using two or more algorithms, more accurate text recognition results can be derived. A device, method, and user terminal may be provided.

An apparatus and method for providing highly accurate text recognition results and an optimal text recognition rate by inputting each candidate text derived using at least two or more algorithms into a pre-trained reinforcement learning model to derive the final text. and a user terminal may be provided.

An apparatus, method, and user terminal can be provided that allow a user to create user rule information for a comparison table created based on each candidate text, and learn the generated user rule information with correct answer data to derive the final text. .

1 is a configuration diagram of a text recognition system according to an embodiment of the present invention.
Figure 2 is a configuration diagram of a user terminal according to an embodiment of the present invention.
Figure 3 is a configuration diagram of a text recognition device according to an embodiment of the present invention.
FIG. 4 is an exemplary diagram illustrating a process of selecting a text recognition engine including an algorithm to be applied to image data according to an embodiment of the present invention.
5A and 5B are exemplary diagrams for explaining a process of deriving final text from image data according to an embodiment of the present invention.
Figure 6 is a flowchart of a method for recognizing text from an image in a text recognition device according to an embodiment of the present invention.
Figure 7 is a flow chart of a method of receiving text recognized from an image in a user terminal according to an embodiment of the present invention.

Below, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case where it is "directly connected," but also the case where it is "electrically connected" with another element in between. . In addition, when a part is said to "include" a certain component, this does not mean excluding other components unless specifically stated to the contrary, but may further include other components, and one or more other features. It should be understood that it does not exclude in advance the presence or addition of numbers, steps, operations, components, parts, or combinations thereof.

In this specification, 'part' includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Additionally, one unit may be realized using two or more pieces of hardware, and two or more units may be realized using one piece of hardware.

In this specification, some of the operations or functions described as being performed by a terminal or device may instead be performed on a server connected to the terminal or device. Likewise, some of the operations or functions described as being performed by the server may also be performed on a terminal or device connected to the server.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the attached drawings.

1 is a configuration diagram of a text recognition system according to an embodiment of the present invention. Referring to FIG. 1, the text recognition system 1 may include a user terminal 110 and a text recognition device 120. The user terminal 110 and the text recognition device 120 are exemplary illustrations of components that can be controlled by the text recognition system 1.

Each component of the text recognition system 1 of FIG. 1 is generally connected through a network. For example, as shown in FIG. 1, the text recognition device 120 may be connected to the user terminal 110 simultaneously or at time intervals.

Network refers to a connection structure that allows information exchange between nodes such as terminals and servers, including Local Area Network (LAN), Wide Area Network (WAN), and World Wide Area Network (WWW). Wide Web), wired and wireless data communication networks, telephone networks, wired and wireless television communication networks, etc. Examples of wireless data communication networks include 3G, 4G, 5G, WIMAX (World Interoperability for Microwave Access), Wi-Fi, Bluetooth communication, infrared communication, ultrasonic communication, Visible Light Communication (VLC), and LiFi. ), etc. are included, but are not limited to this.

The user terminal 110 may select an image to recognize text from among a plurality of images and transmit the selected image to the text recognition device 120.

The user terminal 110 may select at least two text recognition engines including algorithms to be applied to the selected image. For example, the user terminal 110 may receive at least two text recognition engines selected by the user, such as DBN, CNN, DNN, RNN, or GAN. Here, at least two text recognition engines may be learned based on different algorithms.

The user terminal 110 may receive the final text derived through at least two text recognition engines selected from the text recognition device 120.

The user terminal 110 may receive a comparison table generated for each candidate text from the text recognition device 120.

The user terminal 110 may receive user rule information based on the comparison table and transmit the input user rule information to the text recognition device 120. This allows, in the case of a rule or logic that cannot be recognized by the text recognition device 120, a final inspection and confirmation of the text recognized by the user by receiving user rule information. there is.

Through this, cross-checking is performed on the width, height, amount, number of items, etc. related to each candidate text based on the user rule information, so that the text recognition device 120 recognizes the text through OCR and sets the value. A final check can be performed to derive the final text.

The user terminal 110 can check the progress steps for each step until the final text is derived through a progress bar format.

Through this process, the user terminal 110 can check how the text recognition is progressing and by what algorithm, and finally, the user can approve and confirm the result of the final text.

The text recognition device 120 may receive image data including text from the user terminal 110. For example, the text recognition device 120 may receive image data related to receipts, billing statements, business registration certificates, medical expense claim receipts, public institution documents, etc.

The text recognition device 120 may perform preprocessing, such as rotation correction, creasing correction, noise removal, watermark removal, and outer sharpness enhancement, on the received image data.

The text recognition device 120 receives a selection of at least two text recognition engines including algorithms to be applied to image data from the user terminal 110, and selects each candidate recognized from the image data using the at least two selected text recognition engines. Text can be derived.

The text recognition device 120 may generate a comparison table for each candidate text and provide the generated comparison table to the user terminal 110 .

The text recognition device 120 may input each derived candidate text into a pre-trained reinforcement learning model to derive the final text. For example, the text recognition device 120 may perform basic validation on each date, number, English word, and Korean word to derive the final text.

For example, in the case of dates, there are various date formats such as yyyy year, mm month, dd day, yyyy.mm.dd, yy.mm.dd, etc. In this regard, the text recognition device 120 displays the exact year yyyy. The final text can be derived by performing the inspection for mm month, inspection for dd day as the highest priority. At this time, the text recognition device 120 can correct the date format to a numeric format.

If an error occurs in Korean recognition of year/month/day, the text recognition device 120 can apply automatic conversion logic to correct the year, month, and day. For example, the text recognition device 120 may derive the final text by modifying the recognized text into 'Nun' -> 'Yeon', 'Nyan' -> 'Yeon', 'Raat' -> 'Yeon', etc. You can.

For another example, in the case of numbers, 0 to 9 are the default values, and in the case of a comma (,) located in the middle of a number, the text recognition device 120 can directly reflect the format of the original image. For example, the text recognition device 120 can derive the final text by modifying '123.200' -> '123,200'.

For another example, in the case of English text, the text recognition device 120 can recognize A to Z, and English words can be modified using words appearing in Wikipedia, an encyclopedia, etc. to derive the final text.

For another example, in the case of Hangul, the text recognition device 120 performs checks on various terms, word orders, and sentences. To this end, it checks whether accurate text extraction of words excluding particles is prioritized, and then makes corrections. The final text can be derived through this. For example, in the case of a receipt for medical expenses, the text recognition device 120 uses various words that may be generated, such as 'bill statement', 'test fee', 'consultation fee', and 'non-coverage', to apply for medical expenses from the Health Insurance Review and Assessment Service. It can be modified based on the terms of each medical institution and type in the specification.

For another example, in the case of the format of a resident registration number, the format of a business registration certificate, etc., it consists of numbers and '-', and the text recognition device 120 does not perform a separate error check for this, but always recognizes it in a numeric format. This can be modified.

The text recognition device 120 receives user rule information generated based on the comparison table from the user terminal 110, derives the final text based on the received user rule information, and sends the derived final text to the user terminal 110. ) can be provided.

The text recognition device 120 inputs each candidate text and user rule information into a reinforcement learning model to recognize the state, and reinforces it to derive an action that satisfies the reward function from the recognized state. A learning model can be trained.

In other words, by using at least two text recognition engines including algorithms to be applied to image data, the present application has the advantage of being able to obtain higher accuracy than when recognizing text from image data using one text recognition engine. can be provided.

Figure 2 is a configuration diagram of a user terminal according to an embodiment of the present invention. Referring to FIG. 2, the user terminal 110 may include an image selection unit 210, a transmission unit 220, a text recognition engine selection unit 230, a provision unit 240, and a user rule information input unit 250. You can.

The image selection unit 210 may select an image to recognize text from among a plurality of images. For example, when the user terminal 110 accesses a website that provides a text recognition service, the image selection unit 210 selects an image or an image stored in the user terminal 110 through a drag and drop method from the user. You can select an image stored on a cloud server (not shown).

The transmission unit 220 may transmit the selected image to the text recognition device 120.

The text recognition engine selection unit 230 may select at least two text recognition engines including algorithms to be applied to the selected image. For example, the text recognition engine selection unit 230 may select at least two text recognition engines from among a plurality of text recognition engines each composed of algorithms loaded in multi-threads.

The providing unit 240 may receive the final text derived through at least two text recognition engines selected from the text recognition device 120.

The providing unit 240 may further include receiving a comparison table generated for each candidate text from the text recognition device 120.

The user rule information input unit 250 may receive user rule information based on the comparison table. For example, the user rule information input unit 250 receives selections such as 'medical expenses', 'hospital A', and 'telephone number' among the first candidate texts 500 as user rule information from the user, and selects the second candidate text 500 as user rule information. You can select '234-4563' from (510).

The transmission unit 220 may transmit user rule information to the text recognition device 120.

Figure 3 is a configuration diagram of a text recognition device according to an embodiment of the present invention. Referring to FIG. 3, the text recognition device 120 includes a receiving unit 310, a preprocessing unit 320, a text recognition engine selection unit 330, a candidate text deriving unit 340, a comparison table generating unit 350, It may include a final text derivation unit 360, a provision unit 370, and a learning unit 380.

The receiving unit 310 may receive image data including text from the user terminal 110. For example, the receiving unit 310 may receive medical fee receipt image data, receipt fee claim image data, business registration certificate image data, and public institution document image data from the user terminal 110.

The preprocessing unit 320 may perform preprocessing, such as rotation correction, creasing correction, noise removal, watermark removal, and outer sharpness enhancement, on the received image data.

The text recognition engine selection unit 330 may select at least two text recognition engines including algorithms to be applied to image data from the user terminal 110. Here, at least two text recognition engines may each be trained based on different algorithms.

The candidate text derivation unit 340 may derive each candidate text recognized from image data using at least two selected text recognition engines. Here, the candidate text derivation unit 340 may derive candidate texts showing different results depending on the selected text recognition engine.

The final text derivation unit 360 may input each derived candidate text into a pre-trained reinforcement learning model to derive the final text. The process of selecting an algorithm to be applied to image data and deriving each candidate text will be explained in detail with reference to FIG. 4.

FIG. 4 is an exemplary diagram illustrating a process of selecting an algorithm to be applied to image data according to an embodiment of the present invention. Referring to FIG. 4, the text recognition engine selection unit 330 may select at least two text recognition engines including algorithms to be applied to image data. Here, the text recognition engine may be, for example, learned based on an algorithm such as Deep Belief Network (DBN), Convolution Neural Network (CNN), Generative Adversarial Network (GAN), or Recurrent Neural Network (RNN), or at least Two or more text recognition engines may each be trained based on different algorithms.

DBN is a generative graphical model used in machine learning. In deep learning, it refers to a deep neural network composed of multiple layers of latent variables. There are connections between layers, but there are connections between units within the layers. It has the characteristic of having no connection. Due to the nature of DBN as a generative model, it can be used for prior learning. By learning the initial weights through prior learning and then fine-tuning the weights through backpropagation or other discriminant algorithms, the initial values of the previously learned weights are set to the arbitrarily set initial weights. Compared to the value, it becomes closer to the optimal weight, which can improve the performance and speed of the fine-tuning step.

CNN is a type of multilayer perceptrons designed to use minimal preprocessing. It consists of one or several convolution layers and general artificial neural network layers on top of them, including weight and pooling layers. It has the feature of making additional use of them. Due to this structure, it is possible to fully utilize input data in a two-dimensional structure, showing excellent performance in the video and audio fields.

GAN means generative adversarial network, and is composed of two neural networks that are hostile to each other. It includes a data instance generator called a generator and a data reviewer called a discriminator. The generator generates data that is passed to the discriminator, and the discriminator It has the characteristic of determining whether the data is true or false.

RNN is a type of artificial neural network in which hidden nodes are connected by directed edges to form a circular structure. It is composed of a network structure that can accept input and output regardless of the length of the sequence. It has the characteristic of being able to create diverse and flexible structures depending on the type. This RNN is suitable for processing data in which voice, text, etc. appear sequentially.

For example, when the receiver 310 receives 'Hong Gil-dong_A Hospital medical treatment receipt image data' 400 from the user terminal 110, the algorithm selection unit 330 recognizes a plurality of texts from the user terminal 110. Among the engines 410, a first text recognition engine 411 including a 'CNN algorithm' and a second text recognition engine 412 including a 'GAN algorithm' may be selected.

For example, the candidate text derivation unit 340 derives the first candidate text using the first text recognition engine 411 including the 'CNN algorithm', and the second text recognition engine including the 'GAN algorithm'. The second candidate text can be derived using (412).

Returning to FIG. 1, the comparison table generator 350 may generate a comparison table for each candidate text. For example, the comparison table generator 350 may generate a comparison table that compares the first candidate text derived through the first text recognition engine and the second candidate text derived through the second text recognition engine.

The provider 370 may provide the generated comparison table to the user terminal 110 .

The receiving unit 310 may receive user rule information generated based on the comparison table from the user terminal 110. For example, if at least one candidate text '2021.10.03' is recognized as '2021,10,03' due to error recognition, the information recognized as "." among the first candidate text or the second candidate text User rule information using correct answer data can be received.

The final text deriving unit 360 may derive the final text based on the received user rule information.

The provider 370 may provide the derived final text to the user terminal 110.

The learning unit 380 inputs each candidate text and user rule information into a reinforcement learning model to recognize the state, and performs reinforcement learning to derive an action that satisfies the reward function from the recognized state. You can train a model.

5A and 5B are exemplary diagrams for explaining a process of deriving final text from image data according to an embodiment of the present invention.

FIG. 5A is an exemplary diagram illustrating user rule information according to an embodiment of the present invention. Referring to FIG. 5A, the text recognition device 120 derives the first candidate text 500 using a first text recognition engine including a first algorithm, and uses a second text recognition engine including a second algorithm. Let us assume that the second candidate text 510 is derived using

The text recognition device 120 may generate a comparison table that compares the first candidate text 500 and the second candidate text 510 and transmit it to the user terminal 110. At this time, as shown in FIG. 5A, the comparison table may be a table that links and displays each candidate text. That is, the comparison table is a correlation (e.g., It may be a table (displayed in association with a solid line).

The text recognition device 120 may receive user rule information generated based on the comparison table from the user terminal 110. For example, the text recognition device 120 selects 'medical expenses', 'hospital A', and 'telephone number' from the first candidate text 500, and '234-4563' from the second candidate text 510. This selected user rule information can be received.

FIG. 5B is an exemplary diagram illustrating final text according to an embodiment of the present invention. Referring to FIGS. 4 and 5B, when the text recognition device 120 receives ‘Hong Gil-dong_A Hospital medical treatment receipt image data’ 400, the text recognition device 120 receives the final text recognized from the received image data. can be derived.

Figure 6 is a flowchart of a method for recognizing text from an image in a text recognition device according to an embodiment of the present invention. The method of recognizing text from an image in the text recognition device 120 shown in FIG. 6 includes steps processed in time series by the text recognition system 1 according to the embodiment shown in FIGS. 1 to 5B. Therefore, even if the content is omitted below, it also applies to the method of recognizing text from an image in the text recognition device 120 according to the embodiment shown in FIGS. 1 to 5B.

In step S610, the text recognition device 120 may receive image data including text from the user terminal 110.

In step S620, the text recognition device 120 may select at least two text recognition engines including algorithms to be applied to image data from the user terminal 110.

In step S630, the text recognition device 120 may derive each candidate text recognized from the image data using at least two selected text recognition engines.

In step S640, the text recognition device 120 may input each derived candidate text into a pre-trained reinforcement learning model to derive a final text.

In step S650, the text recognition device 120 may provide the derived final text to the user terminal 110.

In the above description, steps S610 to S650 may be further divided into additional steps or combined into fewer steps, depending on the implementation of the present invention. Additionally, some steps may be omitted or the order between steps may be switched as needed.

Figure 7 is a flow chart of a method of receiving text recognized from an image in a user terminal according to an embodiment of the present invention. The method of receiving text recognized from an image in the user terminal 110 shown in FIG. 7 includes steps processed in time series by the text recognition system 1 according to the embodiment shown in FIGS. 1 to 6. . Therefore, even if the content is omitted below, it also applies to the method of receiving text recognized from an image in the user terminal 110 according to the embodiment shown in FIGS. 1 to 6.

In step S710, the user terminal 110 may select an image to recognize text from among a plurality of images.

In step S720, the user terminal 110 may transmit the selected image to the text recognition device 120.

In step S730, the user terminal 110 may select at least two text recognition engines including algorithms to be applied to the selected image.

In step S740, the user terminal 110 may receive the final text derived through at least two text recognition engines selected from the text recognition device 120.

In the above description, steps S710 to S740 may be further divided into additional steps or combined into fewer steps, depending on the implementation of the present invention. Additionally, some steps may be omitted or the order between steps may be switched as needed.

The method of receiving text recognized from an image in a user terminal described with reference to FIGS. 1 to 7 and the method of recognizing text from an image in a text recognition device include a computer program stored on a medium executable by a computer or a computer program executable by a computer. It can also be implemented in the form of a recording medium containing instructions. In addition, the method of receiving text recognized from an image in a user terminal and the method of recognizing text from an image in a text recognition device described with reference to FIGS. 1 to 7 may also be in the form of a computer program stored in a medium executed by a computer. It can be implemented.

Computer-readable media can be any available media that can be accessed by a computer and includes both volatile and non-volatile media, removable and non-removable media. Additionally, computer-readable media may include computer storage media. Computer storage media includes both volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data.

The description of the present invention described above is for illustrative purposes, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential features of the present invention. will be. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. For example, each component described as unitary may be implemented in a distributed manner, and similarly, components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention. do.

110: user terminal
120: Text recognition device
210: Image selection unit
220: Transmission unit
230: Text recognition engine selection unit
240: Provider
250: User rule information input unit
310: receiving unit
320: Preprocessing execution unit
330: Text recognition engine selection unit
340: Candidate text derivation unit
350: Comparison table creation unit
360: Final text derivation unit
370: Provider
380: Learning Department

Claims (16)

In a device for recognizing text from images,
a receiving unit that receives image data including text from a user terminal;
a text recognition engine selection unit that selects at least two text recognition engines including algorithms to be applied to the image data from the user terminal;
a candidate text derivation unit that derives each candidate text recognized from the image data using the selected at least two text recognition engines;
a final text derivation unit that inputs each of the derived candidate texts into a pre-trained reinforcement learning model to derive a final text; and
It includes a providing unit that provides the derived final text to the user terminal,
Further comprising a comparison table generator for generating a comparison table for each candidate text,
The provider provides the generated comparison table to the user terminal,
The receiving unit receives user rule information generated based on the comparison table from the user terminal,
The final text derivation unit derives the final text based on the received user rule information,
The comparison table is,
A text recognition device that associates and displays each candidate text.
According to claim 1,
A text recognition device further comprising a preprocessing unit that performs at least one of rotation correction, creasing correction, noise removal, watermark removal, and outer sharpness enhancement on the received image data.
delete delete According to claim 1,
The reinforcement learning model recognizes a state by inputting each candidate text and the user rule information into the reinforcement learning model, and derives an action that satisfies a reward function from the recognized state. A text recognition device further comprising a learning unit for learning.
In a method of receiving text recognized from an image in a user terminal,
Selecting an image to recognize text from among a plurality of images;
transmitting the selected image to a text recognition device;
selecting at least two text recognition engines including algorithms to be applied to the selected image; and
Comprising: receiving a final text derived through the selected at least two text recognition engines from the text recognition device,
The final text is derived from each candidate text recognized from the image using at least two algorithms selected by the text recognition device,
Each candidate text derived above is input into a pre-trained reinforcement learning model and derived,
receiving a comparison table generated for each candidate text from the text recognition device;
receiving user rule information based on the comparison table; and
It further includes transmitting the input user rule information to the text recognition device,
The comparison table is,
A method of providing text, wherein each candidate text is associated and displayed.
delete delete In a user terminal that receives text recognized from an image,
an image selection unit that selects an image to recognize text from among a plurality of images;
a transmission unit transmitting the selected image to a text recognition device;
an engine selection unit that selects at least two text recognition engines including algorithms to be applied to the selected image; and
A provision unit that receives final text derived from the text recognition device through the selected at least two text recognition engines,
The final text is derived from each candidate text recognized from the image using at least two algorithms selected by the text recognition device,
Each candidate text derived above is input into a pre-trained reinforcement learning model and derived,
The provider receives a comparison table generated for each candidate text from the text recognition device,
Further comprising a user rule information input unit that receives user rule information based on the comparison table,
The transmission unit transmits the input user rule information to the text recognition device,
The comparison table is,
A user terminal that associates and displays each of the candidate texts.
delete delete In a method of recognizing text from an image in a text recognition device,
Receiving image data including text from a user terminal;
Receiving selection of at least two text recognition engines including algorithms to be applied to the image data from the user terminal;
Deriving each candidate text recognized from the image data using the selected at least two text recognition engines;
Inputting each of the derived candidate texts into a pre-trained reinforcement learning model to derive a final text; and
It includes providing the derived final text to the user terminal,
generating a comparison table for each candidate text;
providing the generated comparison table to the user terminal;
Receiving user rule information generated based on the comparison table from the user terminal; and
It further includes deriving the final text based on the received user rule information,
The comparison table is,
A text recognition method that associates and displays each candidate text.
According to claim 12,
A text recognition method further comprising performing at least one preprocessing of rotation correction, creasing correction, noise removal, watermark removal, and outer sharpness enhancement on the received image data.
delete delete According to claim 12,
Each candidate text and the user rule information are input to the reinforcement learning model to recognize a state, and the reinforcement learning is performed to derive an action that satisfies a reward function from the recognized state. A text recognition method further comprising the step of training a model.

Images