WO2022178994A1 - 表格结构识别方法、装置、电子设备及存储介质 - Google Patents

表格结构识别方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2022178994A1
WO2022178994A1 PCT/CN2021/096534 CN2021096534W WO2022178994A1 WO 2022178994 A1 WO2022178994 A1 WO 2022178994A1 CN 2021096534 W CN2021096534 W CN 2021096534W WO 2022178994 A1 WO2022178994 A1 WO 2022178994A1
Authority
WO
WIPO (PCT)
Prior art keywords
relationship
table structure
feature
line
text
Prior art date
Application number
PCT/CN2021/096534
Other languages
English (en)
French (fr)
Inventor
王文浩
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2022178994A1 publication Critical patent/WO2022178994A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/40Document-oriented image-based pattern recognition
    • G06V30/41Analysis of document content
    • G06V30/413Classification of content, e.g. text, photographs or tables
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/40Extraction of image or video features
    • G06V10/44Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/40Document-oriented image-based pattern recognition
    • G06V30/41Analysis of document content
    • G06V30/414Extracting the geometrical structure, e.g. layout tree; Block segmentation, e.g. bounding boxes for graphics or text
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/28Character recognition specially adapted to the type of the alphabet, e.g. Latin alphabet
    • G06V30/287Character recognition specially adapted to the type of the alphabet, e.g. Latin alphabet of Kanji, Hiragana or Katakana characters
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02DCLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present application relates to the technical field of data analysis, and in particular, to a method, apparatus, electronic device, and computer-readable storage medium for identifying a table structure.
  • the traditional table structure recognition is to use the method based on image processing, and use the method of detection or segmentation in the image to recognize and restore the table structure.
  • This method is highly dependent on the image quality. When the image quality is low, the background is complex, and the table color shading is obvious, the detection and recognition effect of the table structure is poor, and the generalization ability is not good.
  • a table structure identification method provided by this application includes:
  • the to-be-identified table page is restored according to the predicted table structure relationship to obtain a table structure.
  • the present application also provides a table structure identification device, the device comprising:
  • a data acquisition module for acquiring a training data set and constructing a label of the training data set
  • a model training module used for training the pre-built original table structure recognition model by using the training data set and the label to obtain a standard table structure recognition model
  • a feature building module used for acquiring the form page to be identified, and constructing the document node feature and the form line feature of the form page to be identified;
  • a table detection module configured to perform table detection and recognition on the document node features and table line features by using the standard table structure recognition model to obtain a predicted table structure relationship
  • a table restoration module configured to perform restoration processing on the to-be-identified table page according to the predicted table structure relationship to obtain a table structure.
  • the present application also provides an electronic device, the electronic device comprising:
  • a processor that executes the instructions stored in the memory to achieve the following steps:
  • the to-be-identified table page is restored according to the predicted table structure relationship to obtain a table structure.
  • the present application also provides a computer-readable storage medium, where the computer-readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the following steps:
  • the to-be-identified table page is restored according to the predicted table structure relationship to obtain a table structure.
  • FIG. 1 is a schematic flowchart of a table structure identification method provided by an embodiment of the present application.
  • FIG. 2 is a functional block diagram of a table structure identification device provided by an embodiment of the present application.
  • FIG. 3 is a schematic structural diagram of an electronic device implementing the method for identifying a table structure according to an embodiment of the present application.
  • the embodiment of the present application provides a table structure identification method.
  • the execution body of the table structure identification method includes, but is not limited to, at least one of electronic devices that can be configured to execute the method provided by the embodiments of the present application, such as a server and a terminal.
  • the method for identifying the table structure can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform.
  • the server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like.
  • the table structure identification method includes:
  • the training data set described in the embodiment of the present application is a picture set containing a table structure, such as a picture converted from a PDF document containing table content.
  • the obtaining of the training data set includes:
  • the page pictures that do not conform to the preset rules are deleted from the page pictures to obtain a training data set.
  • the poppler tool is used to convert each table page into a picture, and the OCR text box is detected and recognized to obtain the OCR parsing result; Rules (for example: the number of tables in the PDF parsing result on the same page is the same as the number of tables in the OCR parsing result, etc.) Compare and verify the PDF parsing result and the OCR parsing result, and delete the images that do not conform to the rules from the multiple images containing tables , to get the training dataset.
  • Rules for example: the number of tables in the PDF parsing result on the same page is the same as the number of tables in the OCR parsing result, etc.
  • the constructing the label of the training data set includes:
  • the embodiment of the present application uses OCR technology to perform text box detection and recognition on the training data set, and the preset relationship conditions include: whether two nodes belong to the same table (table), and whether two nodes belong to the same row (row) and whether the two nodes belong to the same column (col); the adjacency relationship includes a table relationship, a row relationship and a column relationship; the table structure relationship includes a plurality of adjacency relationships.
  • the construction of the adjacency matrix according to the table structure relationship is to regard the cell text box in the table of the training data set as a node in the graph (Graph). If there is an adjacency relationship between any two nodes, the corresponding position in the adjacency matrix is The element is 1, otherwise it is 0. According to the adjacency relationship, three adjacency matrices are constructed as labels for subsequent model training.
  • the original table structure identification model described in the embodiment of the present application is a bert model based on the transformer structure, which can predict the adjacency relationship of each node in the page according to the input feature.
  • the pre-built original table structure recognition model is trained by using the training data set and the label to obtain a standard table structure recognition model, including:
  • the preset loss function may use a mean square error loss function or a cross entropy loss function.
  • the preprocessing of the training data set includes:
  • a text detection result is obtained, and a document node feature is constructed according to the text detection result;
  • the document node features and the table line features are combined to obtain training features.
  • the form page to be identified in the embodiment of the present application may be a picture of an actual bill in the medical field or a converted picture of a document containing a form in the medical field.
  • the to-be-identified form page can be obtained from a pre-built database.
  • the to-be-identified form page may also be obtained from a node of a blockchain.
  • construction of document node features and table line features of the to-be-recognized table page includes:
  • the table line feature is obtained by collecting the position feature of the table line, the text feature of the table line, and the line type feature of the table line.
  • [x1,y1,x2,y2,(x1+x2)/2,(y1+y2)/2,x2-x1,y2-y1] is the position feature
  • (x1,y1) is the upper left corner of the text box
  • the coordinates of the text box (x2, y2) is the coordinates of the lower right corner of the text box, ((x1+x2)/2, (y1+y2)/2) represents the center point of the text box
  • x2-x1 represents the width of the text box
  • y2-y1 represents the height of the text box
  • [num, char, space, other] is the text feature, num, char, space, other respectively represent the frequency of numbers, letters, spaces or other types in the text box calculated according to the content of the text bar
  • the last two bits [0,0] are line type features, which are set to 0 according to the preset line rules, indicating that the node here is a text box, neither horizontal nor vertical.
  • [x1',y1',x2',y2',(x1'+x2')/2,(y1'+y2')/ 2,x2'-x1',y2'-y1'] is the position feature
  • (x1',y1') is the coordinates of the left endpoint of the horizontal line or the endpoint coordinates of the vertical line
  • (x2',y2') is the right endpoint of the horizontal line Coordinates or the coordinates of the lower endpoints of the vertical line
  • ((x1'+x2') /2, (y1'+y2')/2) represent the midpoint of the table line
  • x2'-x1', y2'-y1' represent the horizontal
  • the middle four bits [0,0,0,0] are text features, which are set to 0 according to the preset text conditions, indicating non-text nodes
  • the last two bits [0,1] are line type features
  • the document node feature is the information of the text bar in the form page to be recognized, that is, the content information of the form
  • the table line feature is the information of the table line in the form page to be recognized, that is, the frame of the form information
  • the to-be-recognized table page is represented by constructing document node features and table line features.
  • the S4 includes:
  • the fully connected layer of the standard table structure recognition model is used to perform relationship prediction on the edge features to obtain a predicted table structure relationship, wherein the predicted table structure relationship includes a table relationship, a row relationship and a column relationship.
  • table relationship, row relationship and column relationship in the predicted table structure relationship refer to the table relationship, row relationship and column relationship between any two document nodes in the to-be-identified table page.
  • the standard table structure recognition model described in the embodiment of the present application includes a translation layer, a transformation layer, and a fully connected layer, wherein the translation layer obtains a table page corresponding to the table page to be recognized through operations such as encoding and decoding input features. Each node feature of the relationship) to get the predicted table structure relationship.
  • each of the text boxes is used as a node to construct an undirected graph to obtain a table relationship graph
  • a row relationship diagram and a column relationship diagram are respectively constructed for each of the table sets;
  • the row set and the column set are integrated to obtain a table structure.
  • the maximally connected subgraph of an undirected graph is called the connected component of G.
  • Any connected graph has only one connected component, which is itself.
  • a non-connected undirected graph has multiple connected components.
  • a table can be regarded as a connected graph, and multiple tables in the table page to be identified can be divided by solving the connected components.
  • a clique refers to a complete subgraph of an undirected graph, and a maximal clique is the locally largest clique. If a clique is not contained by any other clique, that is, it is not a proper subset of any other clique, Then the group is called the maximal group of the graph.
  • the maximal clique algorithm is an algorithm for solving all maximal cliques of an undirected graph, specifically including: generating all subgraphs of the undirected graph; judging whether the subgraph is a clique, and deleting the subgraph that is not a clique , get a group; judge whether the group is a maximal group, and delete the group that is not a maximal group to get a maximal group.
  • the row information and column information in a table can be obtained by solving the maximal clique of the row relation graph and the column relation graph respectively.
  • the table structure is restored by a non-image processing method, which can greatly reduce the dependence on the image quality itself.
  • the present application uses the training data set and the label to train a pre-built original table structure recognition model.
  • the training data set is a table data set in the general field and contains a large number of table structures, which can improve the accuracy of the table recognition results.
  • the input of the table structure recognition model is document node features and table line features, and the output is the predicted table structure relationship.
  • the table structure can be restored by restoring the table structure relationship, and the table can be restored by non-image processing methods, which can effectively Avoid the dependence on the image quality itself, reduce the image recognition errors caused by low image quality, complex background or obvious table color shading, and improve the recognition accuracy. Therefore, the table structure identification method, device, electronic device and computer-readable storage medium proposed in the present application can solve the problems of dependence on images and poor table identification effect.
  • FIG. 2 it is a functional block diagram of a table structure identification device provided by an embodiment of the present application.
  • the table structure identification device 100 described in this application can be installed in an electronic device. According to the realized functions, the table structure identification device 100 may include a data acquisition module 101 , a model training module 102 , a feature construction module 103 , a table detection module 104 and a table restoration module 105 .
  • the modules described in this application may also be referred to as units, which refer to a series of computer program segments that can be executed by the processor of an electronic device and can perform fixed functions, and are stored in the memory of the electronic device.
  • each module/unit is as follows:
  • the data acquisition module 101 is configured to acquire a training data set and construct a label of the training data set.
  • the training data set described in the embodiment of the present application is a picture set containing a table structure, such as a picture converted from a PDF document containing table content.
  • the data acquisition module 101 when acquiring the training data set, the data acquisition module 101 specifically performs the following operations:
  • the page pictures that do not conform to the preset rules are deleted from the page pictures to obtain a training data set.
  • the poppler tool is used to convert each table page into a picture, and the OCR text box is detected and recognized to obtain the OCR parsing result; Rules (for example: the number of tables in the PDF parsing result on the same page is the same as the number of tables in the OCR parsing result, etc.) Compare and verify the PDF parsing result and the OCR parsing result, and delete the images that do not conform to the rules from the multiple images containing tables , to get the training dataset.
  • Rules for example: the number of tables in the PDF parsing result on the same page is the same as the number of tables in the OCR parsing result, etc.
  • the data acquisition module 101 specifically performs the following operations:
  • the embodiment of the present application uses OCR technology to perform text box detection and recognition on the training data set, and the preset relationship conditions include: whether two nodes belong to the same table (table), and whether two nodes belong to the same row (row) and whether the two nodes belong to the same column (col); the adjacency relationship includes a table relationship, a row relationship and a column relationship; the table structure relationship includes a plurality of adjacency relationships.
  • the construction of the adjacency matrix according to the table structure relationship is to regard the cell text box in the table of the training data set as a node in the graph (Graph). If there is an adjacency relationship between any two nodes, the corresponding position in the adjacency matrix is The element is 1, otherwise it is 0. According to the adjacency relationship, three adjacency matrices are constructed as labels for subsequent model training.
  • the model training module 102 is configured to use the training data set and the label to train a pre-built original table structure recognition model to obtain a standard table structure recognition model.
  • the original table structure identification model described in the embodiment of the present application is a bert model based on the transformer structure, which can predict the adjacency relationship of each node in the page according to the input feature.
  • model training module 102 is specifically used for:
  • the preset loss function may use a mean square error loss function or a cross entropy loss function.
  • the preprocessing of the training data set includes:
  • a text detection result is obtained, and a document node feature is constructed according to the text detection result;
  • the document node features and the table line features are combined to obtain training features.
  • the feature building module 103 is configured to acquire the form page to be identified, and construct the document node feature and the form line feature of the form page to be identified.
  • the form page to be identified in the embodiment of the present application may be a picture of an actual bill in the medical field or a converted picture of a document containing a form in the medical field.
  • the to-be-identified form page can be obtained from a pre-built database.
  • the to-be-identified form page may also be obtained from a node of a blockchain.
  • the feature building module 103 specifically performs the following operations:
  • the table line feature is obtained by collecting the position feature of the table line, the text feature of the table line, and the line type feature of the table line.
  • [x1,y1,x2,y2,(x1+x2)/2,(y1+y2)/2,x2-x1,y2-y1] is the position feature
  • (x1,y1) is the upper left corner of the text box
  • the coordinates of the text box (x2, y2) is the coordinates of the lower right corner of the text box, ((x1+x2)/2, (y1+y2)/2) represents the center point of the text box
  • x2-x1 represents the width of the text box
  • y2-y1 represents the height of the text box
  • [num, char, space, other] is the text feature, num, char, space, other respectively represent the frequency of numbers, letters, spaces or other types in the text box calculated according to the content of the text bar
  • the last two bits [0,0] are line type features, which are set to 0 according to the preset line rules, indicating that the node here is a text box, neither horizontal nor vertical.
  • [x1',y1',x2',y2',(x1'+x2')/2,(y1'+y2')/ 2,x2'-x1',y2'-y1'] is the position feature
  • (x1',y1') is the coordinates of the left endpoint of the horizontal line or the endpoint coordinates of the vertical line
  • (x2',y2') is the right endpoint of the horizontal line Coordinates or the coordinates of the lower endpoints of the vertical line
  • ((x1'+x2') /2, (y1'+y2')/2) represent the midpoint of the table line
  • x2'-x1', y2'-y1' represent the horizontal
  • the middle four bits [0,0,0,0] are text features, which are set to 0 according to the preset text conditions, indicating non-text nodes
  • the last two bits [0,1] are line type features
  • the document node feature is the information of the text bar in the form page to be recognized, that is, the content information of the form
  • the table line feature is the information of the table line in the form page to be recognized, that is, the frame of the form information
  • the to-be-recognized table page is represented by constructing document node features and table line features.
  • the table detection module 104 is configured to perform table detection and recognition on the document node features and table line features by using the standard table structure recognition model to obtain a predicted table structure relationship.
  • table detection module 104 is specifically used for:
  • the fully connected layer of the standard table structure recognition model is used to perform relationship prediction on the edge features to obtain a predicted table structure relationship, wherein the predicted table structure relationship includes a table relationship, a row relationship and a column relationship.
  • table relationship, row relationship and column relationship in the predicted table structure relationship refer to the table relationship, row relationship and column relationship between any two document nodes in the to-be-identified table page.
  • the standard table structure recognition model described in the embodiment of the present application includes a translation layer, a transformation layer, and a fully connected layer, wherein the translation layer obtains a table page corresponding to the table page to be recognized through operations such as encoding and decoding input features. Each node feature of the relationship) to get the predicted table structure relationship.
  • the table restoration module 105 is configured to perform restoration processing on the to-be-identified table page according to the predicted table structure relationship to obtain a table structure.
  • table restoration module 105 is specifically used for:
  • each of the text boxes is used as a node to construct an undirected graph to obtain a table relationship graph
  • a row relationship diagram and a column relationship diagram are respectively constructed for each of the table sets;
  • the row set and the column set are integrated to obtain a table structure.
  • the maximally connected subgraph of an undirected graph is called the connected component of G.
  • Any connected graph has only one connected component, which is itself.
  • a non-connected undirected graph has multiple connected components.
  • a table can be regarded as a connected graph, and multiple tables in the table page to be identified can be divided by solving the connected components.
  • a clique refers to a complete subgraph of an undirected graph, and a maximal clique is the locally largest clique. If a clique is not contained by any other clique, that is, it is not a proper subset of any other clique, Then the group is called the maximal group of the graph.
  • the maximal clique algorithm is an algorithm for solving all maximal cliques of an undirected graph, specifically including: generating all subgraphs of the undirected graph; judging whether the subgraph is a clique, and deleting the subgraph that is not a clique , get a group; judge whether the group is a maximal group, and delete the group that is not a maximal group to get a maximal group.
  • the row information and column information in a table can be obtained by solving the maximal clique of the row relation graph and the column relation graph respectively.
  • the table structure is restored by a non-image processing method, which can greatly reduce the dependence on the image quality itself.
  • FIG. 3 it is a schematic structural diagram of an electronic device for implementing a table structure identification method provided by an embodiment of the present application.
  • the electronic device 1 may include a processor 10, a memory 11 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a table structure recognition program 12.
  • the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, CD etc.
  • the memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1 .
  • the memory 11 may also be an external storage device of the electronic device 1, such as a pluggable mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital Digital, SD) card, flash memory card (Flash Card), etc.
  • the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device.
  • the memory 11 can not only be used to store application software installed in the electronic device 1 and various types of data, such as the code of the table structure identification program 12, etc., but also can be used to temporarily store data that has been output or will be output.
  • the processor 10 may be composed of integrated circuits, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits packaged with the same function or different functions, including one or more integrated circuits.
  • Central Processing Unit CPU
  • microprocessor digital processing chip
  • graphics processor and combination of various control chips, etc.
  • the processor 10 is the control core (Control Core) of the electronic device. Unit), using various interfaces and lines to connect various components of the entire electronic device, by running or executing programs or modules (such as table structure recognition programs, etc.) stored in the memory 11, and calling the programs stored in the memory 11. data to perform various functions of the electronic device 1 and process data.
  • the bus may be a peripheral component interconnect (PCI for short) bus or an extended industry standard structure (extended). industry standard architecture, referred to as EISA) bus, etc.
  • PCI peripheral component interconnect
  • EISA industry standard architecture
  • the bus can be divided into address bus, data bus, control bus and so on.
  • the bus is configured to implement connection communication between the memory 11 and at least one processor 10 and the like.
  • FIG. 3 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 3 does not constitute a limitation on the electronic device 1, and may include fewer or more components than those shown in the figure. components, or a combination of certain components, or a different arrangement of components.
  • the electronic device 1 may also include a power source (such as a battery) for powering the various components, preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power source can be managed by the power source.
  • the device implements functions such as charge management, discharge management, and power consumption management.
  • the power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components.
  • the electronic device 1 may further include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.
  • the electronic device 1 may also include a network interface, optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
  • a network interface optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
  • the electronic device 1 may further include a user interface, and the user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), optionally, the user interface may also be a standard wired interface or a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, Organic Light Emitting Diode) Touch, etc.
  • the display may also be appropriately called a display screen or a display unit, which is used for displaying information processed in the electronic device 1 and for displaying a visualized user interface.
  • the table structure identification program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions, and when running in the processor 10, it can realize:
  • the to-be-identified table page is restored according to the predicted table structure relationship to obtain a table structure.
  • the modules/units integrated in the electronic device 1 may be stored in a computer-readable storage medium.
  • the computer-readable storage medium may be volatile or non-volatile.
  • the computer-readable storage medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM, Read-Only Memory) Only Memory).
  • the present application also provides a computer-readable storage medium, where the readable storage medium stores a computer program, and when executed by a processor of an electronic device, the computer program can realize:
  • the to-be-identified table page is restored according to the predicted table structure relationship to obtain a table structure.
  • modules described as separate components may or may not be physically separated, and components shown as modules may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.
  • each functional module in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware, or can be implemented in the form of hardware plus software function modules.
  • the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Character Input (AREA)

Abstract

本申请涉及数据分析技术领域,揭露了一种表格结构识别方法,包括:获取训练数据集,并构建标签;利用所述训练数据集和所述标签对原始表格结构识别模型进行训练,得到标准表格结构识别模型;获取待识别表格页面,并构建文档节点特征与表格线特征;利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。此外,本申请还涉及区块链技术,所述待识别表格页面可存储于区块链的节点。本申请还提出一种表格结构识别装置、电子设备以及计算机可读存储介质。本申请可以解决对于图像的依赖性和表格识别效果较差的问题。

Description

表格结构识别方法、装置、电子设备及存储介质
本申请要求于2021年2月24日提交中国专利局、申请号为CN202110206569.3、名称为“表格结构识别方法、装置、电子设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及数据分析技术领域,尤其涉及一种表格结构识别方法、装置、电子设备及计算机可读存储介质。
背景技术
随着大数据时代的到来,如何从海量数据中获取关键、有价值的信息越来越受到重视。如从各大医院、体检机构中患者费用清单、化验单、体检报告等单据中提取信息,可以提高医生后续的诊断效率。文档中的表格结构可以清晰显示出原始文档数据的逻辑与定量关系,很多信息通常以表格的形式呈现,从表格中提取信息前就必不可少的需要先还原表格结构。
技术问题
传统的表格结构识别是采用基于图像处理的方法,采用图像中检测或分割的方法进行表格结构的识别与还原。发明人意识到这种方法高度依赖于图像质量,当图像质量低、背景复杂、表格色彩底纹明显时,表格结构的检测和识别效果较差,同时不具有良好的泛化能力。
技术解决方案
本申请提供的一种表格结构识别方法,包括:
获取训练数据集,并构建所述训练数据集的标签;
利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
本申请还提供一种表格结构识别装置,所述装置包括:
数据获取模块,用于获取训练数据集,并构建所述训练数据集的标签;
模型训练模块,用于利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
特征构建模块,用于获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
表格检测模块,用于利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
表格还原模块,用于根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
本申请还提供一种电子设备,所述电子设备包括:
存储器,存储至少一个指令;及
处理器,执行所述存储器中存储的指令以实现如下步骤:
获取训练数据集,并构建所述训练数据集的标签;
利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
本申请还提供一种计算机可读存储介质,所述计算机可读存储介质中存储有至少一个指令,所述至少一个指令被电子设备中的处理器执行以实现如下步骤:
获取训练数据集,并构建所述训练数据集的标签;
利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
附图说明
图1为本申请一实施例提供的表格结构识别方法的流程示意图;
图2为本申请一实施例提供的表格结构识别装置的功能模块图;
图3为本申请一实施例提供的实现所述表格结构识别方法的电子设备的结构示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
本发明的实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请实施例提供一种表格结构识别方法。所述表格结构识别方法的执行主体包括但不限于服务端、终端等能够被配置为执行本申请实施例提供的该方法的电子设备中的至少一种。换言之,所述表格结构识别方法可以由安装在终端设备或服务端设备的软件或硬件来执行,所述软件可以是区块链平台。所述服务端包括但不限于:单台服务器、服务器集群、云端服务器或云端服务器集群等。
参照图1所示,为本申请一实施例提供的表格结构识别方法的流程示意图。在本实施例中,所述表格结构识别方法包括:
S1、获取训练数据集,并构建所述训练数据集的标签。
本申请实施例中所述训练数据集是含有表格结构的图片集,如含有表格内容的PDF文档转化成的图片。
详细地,所述获取训练数据集,包括:
从网页中爬取多个PDF文档,并对多个所述PDF文档进行解析和筛选,得到多个表格页面;
将每个所述表格页面转化为页面图片,并对所述页面图片进行文字检测与识别,得到识别结果;
根据所述识别结果将所述页面图片中不符合预设规则的页面图片进行删除,得到训练数据集。
例如,从互联网中爬取大量通用领域的PDF文档,并利用pdfplumber库解析PDF文档,得到PDF解析结果。根据解析结果从PDF文档中筛选出含有表格的页面,得到多个表格页面,利用poppler工具将每个表格页面转为图片,并进行OCR文字框检测与识别,得到OCR解析结果;根据预定义的规则(如:同一页PDF解析结果中表格数与OCR解析结果中表格数相同等)将PDF解析结果与OCR解析结果进行对比验证,从所述多个含有表格的图片中删除不符合规则的图片,得到训练数据集。
详细地,所述构建所述训练数据集的标签,包括:
将所述训练数据集进行文本框检测与识别,得到多个文本框;
将所述多个文本框中每个文本框作为一个节点,并根据预设关系条件判断任意两个节点间的邻接关系,得到表格结构关系;
根据所述表格结构关系构造邻接矩阵,得到标签。
进一步地,本申请实施例利用OCR技术对所述训练数据集进行文本框检测与识别,所述预设关系条件包括:两个节点是否属于同一个表格(table)、两个节点是否属于同一行(row)和两个节点是否属于同一列(col);所述邻接关系包括表关系、行关系和列关系;所述表格结构关系包括多个邻接关系。
所述根据所述表格结构关系构造邻接矩阵是将训练数据集的表格内单元格文本框视为是图(Graph)中的节点,若任意两节点间有邻接关系,则邻接矩阵中对应位置处元素为1,否则为0,根据邻接关系构造三个邻接矩阵作为标签,用于后续的模型训练。
S2、利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型。
本申请实施例中所述原始表格结构识别模型是一种基于transformer结构的bert模型,可以根据输入特征预测页面中各节点的邻接关系。
详细地,所述利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型,包括:
对所述训练数据集进行预处理,得到训练特征;
通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵;
根据所述标签和预设的损失函数计算所述关系预测矩阵的损失值;
根据所述损失值对所述原始表格结构识别模型的参数进行调整,并返回上述通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵的步骤,直到所述损失值不再下降,得到标准表格结构识别模型。
其中,所述预设的损失函数可以使用均方误差损失函数或交叉熵损失函数。
进一步地,所述对所述训练数据集进行预处理,包括:
通过对所述训练数据集进行文本框检测与识别,得到文本检测结果,并根据所述文本检测结果构建文档节点特征;
通过对所述训练数据集进行表格线检测,得到检测结果,并根据所述检测构建表格线特征;
将所述文档节点特征和所述表格线特征合并,得到训练特征。
S3、获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征。
本申请实施例中所述待识别表格页面可以是医疗领域的实际票据图片或医疗领域中包含表格的文档转化的图片。所述待识别表格页面可以从预先构建的数据库中获取。为进一步保证所述待识别表格页面的私密性和安全性,所述待识别表格页面还可以从一区块链的节点中获取。
详细地,所述构建所述待识别表格页面的文档节点特征与表格线特征,包括:
对所述待识别表格页面进行文本框检测与识别,得到文本框,其中,所述文本框包括多个文本条和对应的文本框坐标;
根据所述文本框的文本框坐标构建所述文本框的位置特征;
根据所述文本框的文本条构建所述文本框的文本特征;
根据预设线条规则构建所述文本框的线类型特征;
将所述文本框的位置特征、所述文本框的文本特征和所述文本框的线类型特征汇集得到文档节点特征;
对所述待识别表格页面进行表格线检测,得到表格线;
根据所述表格线的端点坐标构建所述表格线的位置特征;
根据预设文本条件构建所述表格线的文本特征;
根据所述表格线的类型构建所述表格线的线类型特征;
将所述表格线的位置特征、所述表格线的文本特征和所述表格线的线类型特征汇集得到表格线特征。
例如:将待识别表格页面检测的每个文本框为一个节点,假设含有N个节点,其文档节点特征D_F(N)=(f(1),f(2),...f(N)),每个节点的特征f(x)=[x1,y1,x2,y2,(x1+x2)/2,(y1+y2)/2,x2-x1,y2-y1,num,char,space, other,0,0]。其中,[x1,y1,x2,y2,(x1+x2)/2,(y1+y2)/2,x2-x1,y2-y1]是位置特征,(x1,y1)为文本框左上角点的坐标,(x2,y2)为文本框右下角点的坐标,((x1+x2)/2,(y1+y2)/2)表示文本框的中心点,x2-x1表示文本框的宽,y2-y1表示文本框的高;[num,char,space,other]是文本特征,num,char,space, other分别表示根据文本条内容计算的文本框中数字、字母、空格或其他类型的频率;最后两位[0,0]是线类型特征,根据预设线条规则设置为0,表示这里的节点是一个文本框,既非横线,也非竖线。
例如:假设待识别表格页面中检测出共有M条横线与竖线,则表格线特征L_F(M) = (f’(1),f’(2),...,f’(M)),其中每条表格线的特征f’(x) = [x1’,y1’,x2’,y2’,(x1’+x2’)/2,(y1’+y2’)/2,x2’-x1’,y2’-y1’, 0,0,0,0,0,1]。其中,[x1’,y1’,x2’,y2’,(x1’+x2’)/2,(y1’+y2’)/ 2,x2’-x1’,y2’-y1’]为位置特征,(x1’,y1’)为横线左端点坐标或竖线上端点坐标,( x2’,y2’)为横线右端点坐标或竖线下端点坐标,((x1’+x2’) /2,(y1’+y2’)/2)表示表格线的中点,x2’-x1’,y2’-y1’分别表示横线和竖线的长度;中间四位[0,0,0,0]为文本特征,根据预设文本条件设置为0,表示非文本节点,最后两位[0,1]为线类型特征,[0,1]表示线类型,[1,0]表示线类型为竖线。
本申请实施例中所述文档节点特征是所述待识别表格页面中文本条的信息,即表格的内容信息,所述表格线特征是所述待识别表格页面中表格线的信息,即表格的框架信息,通过构建文档节点特征与表格线特征来表示所述待识别表格页面。
S4、利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系。
详细地,所述S4,包括:
将所述文档节点特征和所述表格线特征整合,得到输入特征;
利用所述标准表格结构识别模型的翻译层对所述输入特征进行特征提取,得到节点特征;
利用所述标准表格结构识别模型的变换层对所述节点特征输入进行双线性变换,得到边特征;
利用所述标准表格结构识别模型的全连接层对所述边特征进行关系预测,得到预测表格结构关系,其中,所述预测表格结构关系包括表关系、行关系和列关系。
进一步地,所述预测表格结构关系中的表关系、行关系和列关系是指所述待识别表格页面中任意两个文档节点之间的表关系、行关系和列关系。
可选地,本申请实施例中所述标准表格结构识别模型包括翻译层、变换层和全连接层,其中,所述翻译层通过对输入特征进行编码、解码等操作得到与待识别表格页面对应的各节点特征;所述变换层通过进行线性变换得到任意两节点之间的特征,即边特征;所述全连接层则是通过预设的参数计算得任意两节点之间的关系特征(邻接关系),得到预测表格结构关系。
S5、根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
详细地,所述根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构,包括:
对所述待识别表格页面进行文本框检测与识别,得到多个文本框;
根据所述预测表格结构关系中的表关系将每个所述文本框作为节点,构建无向图,得到表关系图;
通过求解所述表关系图的连通分量将所述节点划分为多个表格集;
根据所述预测表格结构关系的行关系和列关系对每个所述表格集分别构建行关系图和列关系图;
利用极大团算法求解所述行关系图中的行极大团,并按照行极大团的纵坐标从大到小进行排序,得到行集合;
利用极大团算法求解所述列关系图中的列极大团,并按照列极大团的横坐标从小到大进行排序,得到列集合;
将所述行集合和所述列集合进行整合,得到表格结构。
进一步地,无向图的极大连通子图称为G的连通分量,任何连通图的连通分量只有一个,即是其自身,非连通的无向图有多个连通分量。一个表格可以视为一个连通图,通过求解连通分量可以划分出所述待识别表格页面中的多个表格。
可选地,团是指一个无向图的完全子图,极大团就是在局部上最大的团,如果一个团不被其他任一团所包含,即它不是其他任一团的真子集,则称该团为图的极大团。
所述极大团算法是用于求解一个无向图的所有极大团的算法,具体包括:生成无向图的所有子图;判断所述子图是不是团,并删除不是团的子图,得到团;判断所述团是不是极大团,并删除不是极大团的团,得到极大团。
本申请实施例通过分别求解行关系图和列关系图的极大团可以得到一个表格中的行信息和列信息。
本申请实施例是通过非图像处理方法实现表格结构还原的,可以大大减少对图像质量本身的依赖。
本申请利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,所述训练数据集是通用领域的表格数据集,包含大量的表格结构,可以提高表格识别结果的准确性;同时,表格结构识别模型的输入为文档节点特征与表格线特征,输出为预测的表格结构关系,通过表格结构关系进行还原处理来还原表格结构,通过非图像处理方法来还原表格,可以有效避免对图像质量本身的依赖性,减少因图像质量低、背景复杂或表格色彩底纹明显造成的图像识别误差,提高识别准确度。因此本申请提出的表格结构识别方法、装置、电子设备及计算机可读存储介质,可以解决对于图像的依赖性和表格识别效果较差的问题。
如图2所示,是本申请一实施例提供的表格结构识别装置的功能模块图。
本申请所述表格结构识别装置100可以安装于电子设备中。根据实现的功能,所述表格结构识别装置100可以包括数据获取模块101、模型训练模块102、特征构建模块103、表格检测模块104及表格还原模块105。本申请所述模块也可以称之为单元,是指一种能够被电子设备处理器所执行,并且能够完成固定功能的一系列计算机程序段,其存储在电子设备的存储器中。
在本实施例中,关于各模块/单元的功能如下:
所述数据获取模块101,用于获取训练数据集,并构建所述训练数据集的标签。
本申请实施例中所述训练数据集是含有表格结构的图片集,如含有表格内容的PDF文档转化成的图片。
详细地,在获取训练数据集时,所述数据获取模块101具体执行下述操作:
从网页中爬取多个PDF文档,并对多个所述PDF文档进行解析和筛选,得到多个表格页面;
将每个所述表格页面转化为页面图片,并对所述页面图片进行文字检测与识别,得到识别结果;
根据所述识别结果将所述页面图片中不符合预设规则的页面图片进行删除,得到训练数据集。
例如,从互联网中爬取大量通用领域的PDF文档,并利用pdfplumber库解析PDF文档,得到PDF解析结果。根据解析结果从PDF文档中筛选出含有表格的页面,得到多个表格页面,利用poppler工具将每个表格页面转为图片,并进行OCR文字框检测与识别,得到OCR解析结果;根据预定义的规则(如:同一页PDF解析结果中表格数与OCR解析结果中表格数相同等)将PDF解析结果与OCR解析结果进行对比验证,从所述多个含有表格的图片中删除不符合规则的图片,得到训练数据集。
详细地,在构建所述训练数据集的标签时,所述数据获取模块101具体执行下述操作:
将所述训练数据集进行文本框检测与识别,得到多个文本框;
将所述多个文本框中每个文本框作为一个节点,并根据预设关系条件判断任意两个节点间的邻接关系,得到表格结构关系;
根据所述表格结构关系构造邻接矩阵,得到标签。
进一步地,本申请实施例利用OCR技术对所述训练数据集进行文本框检测与识别,所述预设关系条件包括:两个节点是否属于同一个表格(table)、两个节点是否属于同一行(row)和两个节点是否属于同一列(col);所述邻接关系包括表关系、行关系和列关系;所述表格结构关系包括多个邻接关系。
所述根据所述表格结构关系构造邻接矩阵是将训练数据集的表格内单元格文本框视为是图(Graph)中的节点,若任意两节点间有邻接关系,则邻接矩阵中对应位置处元素为1,否则为0,根据邻接关系构造三个邻接矩阵作为标签,用于后续的模型训练。
所述模型训练模块102,用于利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型。
本申请实施例中所述原始表格结构识别模型是一种基于transformer结构的bert模型,可以根据输入特征预测页面中各节点的邻接关系。
详细地,所述模型训练模块102具体用于:
对所述训练数据集进行预处理,得到训练特征;
通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵;
根据所述标签和预设的损失函数计算所述关系预测矩阵的损失值;
根据所述损失值对所述原始表格结构识别模型的参数进行调整,并返回上述通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵的步骤,直到所述损失值不再下降,得到标准表格结构识别模型。
其中,所述预设的损失函数可以使用均方误差损失函数或交叉熵损失函数。
进一步地,所述对所述训练数据集进行预处理,包括:
通过对所述训练数据集进行文本框检测与识别,得到文本检测结果,并根据所述文本检测结果构建文档节点特征;
通过对所述训练数据集进行表格线检测,得到检测结果,并根据所述检测构建表格线特征;
将所述文档节点特征和所述表格线特征合并,得到训练特征。
所述特征构建模块103,用于获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征。
本申请实施例中所述待识别表格页面可以是医疗领域的实际票据图片或医疗领域中包含表格的文档转化的图片。所述待识别表格页面可以从预先构建的数据库中获取。为进一步保证所述待识别表格页面的私密性和安全性,所述待识别表格页面还可以从一区块链的节点中获取。
详细地,在构建所述待识别表格页面的文档节点特征与表格线特征时,所述特征构建模块103具体执行下述操作:
对所述待识别表格页面进行文本框检测与识别,得到文本框,其中,所述文本框包括多个文本条和对应的文本框坐标;
根据所述文本框的文本框坐标构建所述文本框的位置特征;
根据所述文本框的文本条构建所述文本框的文本特征;
根据预设线条规则构建所述文本框的线类型特征;
将所述文本框的位置特征、所述文本框的文本特征和所述文本框的线类型特征汇集得到文档节点特征;
对所述待识别表格页面进行表格线检测,得到表格线;
根据所述表格线的端点坐标构建所述表格线的位置特征;
根据预设文本条件构建所述表格线的文本特征;
根据所述表格线的类型构建所述表格线的线类型特征;
将所述表格线的位置特征、所述表格线的文本特征和所述表格线的线类型特征汇集得到表格线特征。
例如:将待识别表格页面检测的每个文本框为一个节点,假设含有N个节点,其文档节点特征D_F(N)=(f(1),f(2),...f(N)),每个节点的特征f(x)=[x1,y1,x2,y2,(x1+x2)/2,(y1+y2)/2,x2-x1,y2-y1,num,char,space, other,0,0]。其中,[x1,y1,x2,y2,(x1+x2)/2,(y1+y2)/2,x2-x1,y2-y1]是位置特征,(x1,y1)为文本框左上角点的坐标,(x2,y2)为文本框右下角点的坐标,((x1+x2)/2,(y1+y2)/2)表示文本框的中心点,x2-x1表示文本框的宽,y2-y1表示文本框的高;[num,char,space,other]是文本特征,num,char,space, other分别表示根据文本条内容计算的文本框中数字、字母、空格或其他类型的频率;最后两位[0,0]是线类型特征,根据预设线条规则设置为0,表示这里的节点是一个文本框,既非横线,也非竖线。
例如:假设待识别表格页面中检测出共有M条横线与竖线,则表格线特征L_F(M) = (f’(1),f’(2),...,f’(M)),其中每条表格线的特征f’(x) = [x1’,y1’,x2’,y2’,(x1’+x2’)/2,(y1’+y2’)/2,x2’-x1’,y2’-y1’, 0,0,0,0,0,1]。其中,[x1’,y1’,x2’,y2’,(x1’+x2’)/2,(y1’+y2’)/ 2,x2’-x1’,y2’-y1’]为位置特征,(x1’,y1’)为横线左端点坐标或竖线上端点坐标,( x2’,y2’)为横线右端点坐标或竖线下端点坐标,((x1’+x2’) /2,(y1’+y2’)/2)表示表格线的中点,x2’-x1’,y2’-y1’分别表示横线和竖线的长度;中间四位[0,0,0,0]为文本特征,根据预设文本条件设置为0,表示非文本节点,最后两位[0,1]为线类型特征,[0,1]表示线类型,[1,0]表示线类型为竖线。
本申请实施例中所述文档节点特征是所述待识别表格页面中文本条的信息,即表格的内容信息,所述表格线特征是所述待识别表格页面中表格线的信息,即表格的框架信息,通过构建文档节点特征与表格线特征来表示所述待识别表格页面。
所述表格检测模块104,用于利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系。
详细地,所述表格检测模块104具体用于:
将所述文档节点特征和所述表格线特征整合,得到输入特征;
利用所述标准表格结构识别模型的翻译层对所述输入特征进行特征提取,得到节点特征;
利用所述标准表格结构识别模型的变换层对所述节点特征输入进行双线性变换,得到边特征;
利用所述标准表格结构识别模型的全连接层对所述边特征进行关系预测,得到预测表格结构关系,其中,所述预测表格结构关系包括表关系、行关系和列关系。
进一步地,所述预测表格结构关系中的表关系、行关系和列关系是指所述待识别表格页面中任意两个文档节点之间的表关系、行关系和列关系。
可选地,本申请实施例中所述标准表格结构识别模型包括翻译层、变换层和全连接层,其中,所述翻译层通过对输入特征进行编码、解码等操作得到与待识别表格页面对应的各节点特征;所述变换层通过进行线性变换得到任意两节点之间的特征,即边特征;所述全连接层则是通过预设的参数计算得任意两节点之间的关系特征(邻接关系),得到预测表格结构关系。
所述表格还原模块105,用于根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
详细地,所述表格还原模块105具体用于:
对所述待识别表格页面进行文本框检测与识别,得到多个文本框;
根据所述预测表格结构关系中的表关系将每个所述文本框作为节点,构建无向图,得到表关系图;
通过求解所述表关系图的连通分量将所述节点划分为多个表格集;
根据所述预测表格结构关系的行关系和列关系对每个所述表格集分别构建行关系图和列关系图;
利用极大团算法求解所述行关系图中的行极大团,并按照行极大团的纵坐标从大到小进行排序,得到行集合;
利用极大团算法求解所述列关系图中的列极大团,并按照列极大团的横坐标从小到大进行排序,得到列集合;
将所述行集合和所述列集合进行整合,得到表格结构。
进一步地,无向图的极大连通子图称为G的连通分量,任何连通图的连通分量只有一个,即是其自身,非连通的无向图有多个连通分量。一个表格可以视为一个连通图,通过求解连通分量可以划分出所述待识别表格页面中的多个表格。
可选地,团是指一个无向图的完全子图,极大团就是在局部上最大的团,如果一个团不被其他任一团所包含,即它不是其他任一团的真子集,则称该团为图的极大团。
所述极大团算法是用于求解一个无向图的所有极大团的算法,具体包括:生成无向图的所有子图;判断所述子图是不是团,并删除不是团的子图,得到团;判断所述团是不是极大团,并删除不是极大团的团,得到极大团。
本申请实施例通过分别求解行关系图和列关系图的极大团可以得到一个表格中的行信息和列信息。
本申请实施例是通过非图像处理方法实现表格结构还原的,可以大大减少对图像质量本身的依赖。
如图3所示,是本申请一实施例提供的实现表格结构识别方法的电子设备的结构示意图。
所述电子设备1可以包括处理器10、存储器11和总线,还可以包括存储在所述存储器11中并可在所述处理器10上运行的计算机程序,如表格结构识别程序12。
其中,所述存储器11至少包括一种类型的可读存储介质,所述可读存储介质包括闪存、移动硬盘、多媒体卡、卡型存储器(例如:SD或DX存储器等)、磁性存储器、磁盘、光盘等。所述存储器11在一些实施例中可以是电子设备1的内部存储单元,例如该电子设备1的移动硬盘。所述存储器11在另一些实施例中也可以是电子设备1的外部存储设备,例如电子设备1上配备的插接式移动硬盘、智能存储卡(Smart Media Card, SMC)、安全数字(Secure Digital, SD)卡、闪存卡(Flash Card)等。进一步地,所述存储器11还可以既包括电子设备1的内部存储单元也包括外部存储设备。所述存储器11不仅可以用于存储安装于电子设备1的应用软件及各类数据,例如表格结构识别程序12的代码等,还可以用于暂时地存储已经输出或者将要输出的数据。
所述处理器10在一些实施例中可以由集成电路组成,例如可以由单个封装的集成电路所组成,也可以是由多个相同功能或不同功能封装的集成电路所组成,包括一个或者多个中央处理器(Central Processing unit,CPU)、微处理器、数字处理芯片、图形处理器及各种控制芯片的组合等。所述处理器10是所述电子设备的控制核心(Control Unit),利用各种接口和线路连接整个电子设备的各个部件,通过运行或执行存储在所述存储器11内的程序或者模块(例如表格结构识别程序等),以及调用存储在所述存储器11内的数据,以执行电子设备1的各种功能和处理数据。
所述总线可以是外设部件互连标准(peripheral component interconnect,简称PCI)总线或扩展工业标准结构(extended industry standard architecture,简称EISA)总线等。该总线可以分为地址总线、数据总线、控制总线等。所述总线被设置为实现所述存储器11以及至少一个处理器10等之间的连接通信。
图3仅示出了具有部件的电子设备,本领域技术人员可以理解的是,图3示出的结构并不构成对所述电子设备1的限定,可以包括比图示更少或者更多的部件,或者组合某些部件,或者不同的部件布置。
例如,尽管未示出,所述电子设备1还可以包括给各个部件供电的电源(比如电池),优选地,电源可以通过电源管理装置与所述至少一个处理器10逻辑相连,从而通过电源管理装置实现充电管理、放电管理、以及功耗管理等功能。电源还可以包括一个或一个以上的直流或交流电源、再充电装置、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。所述电子设备1还可以包括多种传感器、蓝牙模块、Wi-Fi模块等,在此不再赘述。
进一步地,所述电子设备1还可以包括网络接口,可选地,所述网络接口可以包括有线接口和/或无线接口(如WI-FI接口、蓝牙接口等),通常用于在该电子设备1与其他电子设备之间建立通信连接。
可选地,该电子设备1还可以包括用户接口,用户接口可以是显示器(Display)、输入单元(比如键盘(Keyboard)),可选地,用户接口还可以是标准的有线接口、无线接口。可选地,在一些实施例中,显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。其中,显示器也可以适当的称为显示屏或显示单元,用于显示在电子设备1中处理的信息以及用于显示可视化的用户界面。
应该了解,所述实施例仅为说明之用,在专利申请范围上并不受此结构的限制。
所述电子设备1中的所述存储器11存储的表格结构识别程序12是多个指令的组合,在所述处理器10中运行时,可以实现:
获取训练数据集,并构建所述训练数据集的标签;
利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
具体地,所述处理器10对上述指令的具体实现方法可参考图1至图3对应实施例中相关步骤的描述,在此不赘述。
进一步地,所述电子设备1集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读存储介质中。所述计算机可读存储介质可以是易失性的,也可以是非易失性的。例如,所述计算机可读存储介质可以包括:能够携带所述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM,Read-Only Memory)。
本申请还提供一种计算机可读存储介质,所述可读存储介质存储有计算机程序,所述计算机程序在被电子设备的处理器所执行时,可以实现:
获取训练数据集,并构建所述训练数据集的标签;
利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
在本申请所提供的几个实施例中,应该理解到,所揭露的设备,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述模块的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式。
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的,作为模块显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能模块可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用硬件加软件功能模块的形式实现。
对于本领域技术人员而言,显然本申请不限于上述示范性实施例的细节,而且在不背离本申请的精神或基本特征的情况下,能够以其他的具体形式实现本申请。
因此,无论从哪一点来看,均应将实施例看作是示范性的,而且是非限制性的,本申请的范围由所附权利要求而不是上述说明限定,因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本申请内。不应将权利要求中的任何附关联图标记视为限制所涉及的权利要求。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
此外,显然“包括”一词不排除其他单元或步骤,单数不排除复数。系统权利要求中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第二等词语用来表示名称,而并不表示任何特定的顺序。
最后应说明的是,以上实施例仅用以说明本申请的技术方案而非限制,尽管参照较佳实施例对本申请进行了详细说明,本领域的普通技术人员应当理解,可以对本申请的技术方案进行修改或等同替换,而不脱离本申请技术方案的精神和范围。

Claims (20)

  1. 一种表格结构识别方法,其中,所述方法包括:
    获取训练数据集,并构建所述训练数据集的标签;
    利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
    获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
    利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
    根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
  2. 如权利要求1所述的表格结构识别方法,其中,所述获取训练数据集,包括:
    从网页中爬取多个PDF文档,并对多个所述PDF文档进行解析和筛选,得到多个表格页面;
    将每个所述表格页面转化为页面图片,并对所述页面图片进行文字检测与识别,得到识别结果;
    根据所述识别结果将所述页面图片中不符合预设规则的页面图片进行删除,得到训练数据集。
  3. 如权利要求1或2所述的表格结构识别方法,其中,所述构建所述训练数据集的标签,包括:
    将所述训练数据集进行文本框检测与识别,得到多个文本框;
    将所述多个文本框中每个文本框作为一个节点,并根据预设关系条件判断任意两个节点间的邻接关系,得到表格结构关系;
    根据所述表格结构关系构造邻接矩阵,得到标签。
  4. 如权利要求3所述的表格结构识别方法,其中,所述利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型,包括:
    对所述训练数据集进行预处理,得到训练特征;
    通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵;
    根据所述标签和预设的损失函数计算所述关系预测矩阵的损失值;
    根据所述损失值对所述原始表格结构识别模型的参数进行调整,并返回上述通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵的步骤,直到所述损失值不再下降,得到标准表格结构识别模型。
  5. 如权利要求1所述的表格结构识别方法,其中,所述构建所述待识别表格页面的文档节点特征与表格线特征,包括:
    对所述待识别表格页面进行文本框检测与识别,得到文本框,其中,所述文本框包括多个文本条和对应的文本框坐标;
    根据所述文本框的文本框坐标构建所述文本框的位置特征;
    根据所述文本框的文本条构建所述文本框的文本特征;
    根据预设线条规则构建所述文本框的线类型特征;
    将所述文本框的位置特征、所述文本框的文本特征和所述文本框的线类型特征汇集得到文档节点特征;
    对所述待识别表格页面进行表格线检测,得到表格线;
    根据所述表格线的端点坐标构建所述表格线的位置特征;
    根据预设文本条件构建所述表格线的文本特征;
    根据所述表格线的类型构建所述表格线的线类型特征;
    将所述表格线的位置特征、所述表格线的文本特征和所述表格线的线类型特征汇集得到表格线特征。
  6. 如权利要求1所述的表格结构识别方法,其中,所述利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系,包括:
    将所述文档节点特征和所述表格线特征整合,得到输入特征;
    利用所述标准表格结构识别模型的翻译层对所述输入特征进行特征提取,得到节点特征;
    利用所述标准表格结构识别模型的变换层对所述节点特征进行双线性变换,得到边特征;
    利用所述标准表格结构识别模型的全连接层对所述边特征进行关系预测,得到预测表格结构关系,其中,所述预测表格结构关系包括表关系、行关系和列关系。
  7. 如权利要求6所述的表格结构识别方法,其中,所述根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构,包括:
    对所述待识别表格页面进行文本框检测与识别,得到多个文本框;
    根据所述预测表格结构关系中的表关系将每个所述文本框作为节点,构建无向图,得到表关系图;
    通过求解所述表关系图的连通分量将所述节点划分为多个表格集;
    根据所述预测表格结构关系的行关系和列关系对每个所述表格集分别构建行关系图和列关系图;
    利用极大团算法求解所述行关系图中的行极大团,并按照行极大团的纵坐标从大到小进行排序,得到行集合;
    利用极大团算法求解所述列关系图中的列极大团,并按照列极大团的横坐标从小到大进行排序,得到列集合;
    将所述行集合和所述列集合进行整合,得到表格结构。
  8. 一种表格结构识别装置,其中,所述装置包括:
    数据获取模块,用于获取训练数据集,并构建所述训练数据集的标签;
    模型训练模块,用于利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
    特征构建模块,用于获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
    表格检测模块,用于利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
    表格还原模块,用于根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
  9. 一种电子设备,其中,所述电子设备包括:
    至少一个处理器;以及,
    与所述至少一个处理器通信连接的存储器;其中,
    所述存储器存储有可被所述至少一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器能够执行如下步骤:
    获取训练数据集,并构建所述训练数据集的标签;
    利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
    获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
    利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
    根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
  10. 如权利要求9所述的电子设备,其中,所述获取训练数据集,包括:
    从网页中爬取多个PDF文档,并对多个所述PDF文档进行解析和筛选,得到多个表格页面;
    将每个所述表格页面转化为页面图片,并对所述页面图片进行文字检测与识别,得到识别结果;
    根据所述识别结果将所述页面图片中不符合预设规则的页面图片进行删除,得到训练数据集。
  11. 如权利要求9或10所述的电子设备,其中,所述构建所述训练数据集的标签,包括:
    将所述训练数据集进行文本框检测与识别,得到多个文本框;
    将所述多个文本框中每个文本框作为一个节点,并根据预设关系条件判断任意两个节点间的邻接关系,得到表格结构关系;
    根据所述表格结构关系构造邻接矩阵,得到标签。
  12. 如权利要求11所述的电子设备,其中,所述利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型,包括:
    对所述训练数据集进行预处理,得到训练特征;
    通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵;
    根据所述标签和预设的损失函数计算所述关系预测矩阵的损失值;
    根据所述损失值对所述原始表格结构识别模型的参数进行调整,并返回上述通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵的步骤,直到所述损失值不再下降,得到标准表格结构识别模型。
  13. 如权利要求9所述的电子设备,其中,所述构建所述待识别表格页面的文档节点特征与表格线特征,包括:
    对所述待识别表格页面进行文本框检测与识别,得到文本框,其中,所述文本框包括多个文本条和对应的文本框坐标;
    根据所述文本框的文本框坐标构建所述文本框的位置特征;
    根据所述文本框的文本条构建所述文本框的文本特征;
    根据预设线条规则构建所述文本框的线类型特征;
    将所述文本框的位置特征、所述文本框的文本特征和所述文本框的线类型特征汇集得到文档节点特征;
    对所述待识别表格页面进行表格线检测,得到表格线;
    根据所述表格线的端点坐标构建所述表格线的位置特征;
    根据预设文本条件构建所述表格线的文本特征;
    根据所述表格线的类型构建所述表格线的线类型特征;
    将所述表格线的位置特征、所述表格线的文本特征和所述表格线的线类型特征汇集得到表格线特征。
  14. 如权利要求9所述的电子设备,其中,所述利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系,包括:
    将所述文档节点特征和所述表格线特征整合,得到输入特征;
    利用所述标准表格结构识别模型的翻译层对所述输入特征进行特征提取,得到节点特征;
    利用所述标准表格结构识别模型的变换层对所述节点特征进行双线性变换,得到边特征;
    利用所述标准表格结构识别模型的全连接层对所述边特征进行关系预测,得到预测表格结构关系,其中,所述预测表格结构关系包括表关系、行关系和列关系。
  15. 如权利要求14所述的电子设备,其中,所述根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构,包括:
    对所述待识别表格页面进行文本框检测与识别,得到多个文本框;
    根据所述预测表格结构关系中的表关系将每个所述文本框作为节点,构建无向图,得到表关系图;
    通过求解所述表关系图的连通分量将所述节点划分为多个表格集;
    根据所述预测表格结构关系的行关系和列关系对每个所述表格集分别构建行关系图和列关系图;
    利用极大团算法求解所述行关系图中的行极大团,并按照行极大团的纵坐标从大到小进行排序,得到行集合;
    利用极大团算法求解所述列关系图中的列极大团,并按照列极大团的横坐标从小到大进行排序,得到列集合;
    将所述行集合和所述列集合进行整合,得到表格结构。
  16. 一种计算机可读存储介质,存储有计算机程序,其中,所述计算机程序被处理器执行时实现如下步骤:
    获取训练数据集,并构建所述训练数据集的标签;
    利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型;
    获取待识别表格页面,并构建所述待识别表格页面的文档节点特征与表格线特征;
    利用所述标准表格结构识别模型对所述文档节点特征与表格线特征进行表格检测与识别,得到预测表格结构关系;
    根据所述预测表格结构关系对所述待识别表格页面进行还原处理,得到表格结构。
  17. 如权利要求16所述的计算机可读存储介质,其中,所述获取训练数据集,包括:
    从网页中爬取多个PDF文档,并对多个所述PDF文档进行解析和筛选,得到多个表格页面;
    将每个所述表格页面转化为页面图片,并对所述页面图片进行文字检测与识别,得到识别结果;
    根据所述识别结果将所述页面图片中不符合预设规则的页面图片进行删除,得到训练数据集。
  18. 如权利要求16或17所述的计算机可读存储介质,其中,所述构建所述训练数据集的标签,包括:
    将所述训练数据集进行文本框检测与识别,得到多个文本框;
    将所述多个文本框中每个文本框作为一个节点,并根据预设关系条件判断任意两个节点间的邻接关系,得到表格结构关系;
    根据所述表格结构关系构造邻接矩阵,得到标签。
  19. 如权利要求18所述的计算机可读存储介质,其中,所述利用所述训练数据集和所述标签对预构建的原始表格结构识别模型进行训练,得到标准表格结构识别模型,包括:
    对所述训练数据集进行预处理,得到训练特征;
    通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵;
    根据所述标签和预设的损失函数计算所述关系预测矩阵的损失值;
    根据所述损失值对所述原始表格结构识别模型的参数进行调整,并返回上述通过所述原始表格结构识别模型对所述训练特征进行表格识别,得到关系预测矩阵的步骤,直到所述损失值不再下降,得到标准表格结构识别模型。
  20. 如权利要求16所述的计算机可读存储介质,其中,所述构建所述待识别表格页面的文档节点特征与表格线特征,包括:
    对所述待识别表格页面进行文本框检测与识别,得到文本框,其中,所述文本框包括多个文本条和对应的文本框坐标;
    根据所述文本框的文本框坐标构建所述文本框的位置特征;
    根据所述文本框的文本条构建所述文本框的文本特征;
    根据预设线条规则构建所述文本框的线类型特征;
    将所述文本框的位置特征、所述文本框的文本特征和所述文本框的线类型特征汇集得到文档节点特征;
    对所述待识别表格页面进行表格线检测,得到表格线;
    根据所述表格线的端点坐标构建所述表格线的位置特征;
    根据预设文本条件构建所述表格线的文本特征;
    根据所述表格线的类型构建所述表格线的线类型特征;
    将所述表格线的位置特征、所述表格线的文本特征和所述表格线的线类型特征汇集得到表格线特征。
PCT/CN2021/096534 2021-02-24 2021-05-27 表格结构识别方法、装置、电子设备及存储介质 WO2022178994A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110206569.3 2021-02-24
CN202110206569.3A CN112949443B (zh) 2021-02-24 2021-02-24 表格结构识别方法、装置、电子设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022178994A1 true WO2022178994A1 (zh) 2022-09-01

Family

ID=76245817

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/096534 WO2022178994A1 (zh) 2021-02-24 2021-05-27 表格结构识别方法、装置、电子设备及存储介质

Country Status (2)

Country Link
CN (1) CN112949443B (zh)
WO (1) WO2022178994A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118351543A (zh) * 2024-06-18 2024-07-16 南昌大学第一附属医院 一种医疗检验单的数据信息提取分析方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113610043B (zh) * 2021-08-19 2024-09-27 海默潘多拉数据科技(深圳)有限公司 一种工业图纸表格结构化识别方法及系统
CN113762158A (zh) * 2021-09-08 2021-12-07 平安资产管理有限责任公司 无边框表格复原模型训练方法、装置、计算机设备和介质
CN113849552B (zh) * 2021-09-27 2024-05-31 中国平安财产保险股份有限公司 结构化数据转换方法、装置、电子设备及介质
CN113869017B (zh) * 2021-09-30 2024-08-16 平安科技(深圳)有限公司 基于人工智能的表格图像重构方法、装置、设备及介质
CN113887441B (zh) * 2021-09-30 2024-09-10 平安银行股份有限公司 一种表格数据处理方法、装置、设备及存储介质
CN115116060B (zh) * 2022-08-25 2023-01-24 深圳前海环融联易信息科技服务有限公司 键值文件处理方法、装置、设备、介质
CN116127927B (zh) * 2023-04-04 2023-06-16 北京智麟科技有限公司 一种网页表格转pdf文件的方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110334585A (zh) * 2019-05-22 2019-10-15 平安科技(深圳)有限公司 表格识别方法、装置、计算机设备和存储介质
US20200151443A1 (en) * 2018-11-09 2020-05-14 Microsoft Technology Licensing, Llc Supervised ocr training for custom forms
CN111382717A (zh) * 2020-03-17 2020-07-07 腾讯科技(深圳)有限公司 一种表格识别方法、装置和计算机可读存储介质
CN111860257A (zh) * 2020-07-10 2020-10-30 上海交通大学 融合多种文本特征及几何信息的表格识别方法及系统
CN112381010A (zh) * 2020-11-17 2021-02-19 深圳壹账通智能科技有限公司 表格结构的还原方法、系统、计算机设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200151443A1 (en) * 2018-11-09 2020-05-14 Microsoft Technology Licensing, Llc Supervised ocr training for custom forms
CN110334585A (zh) * 2019-05-22 2019-10-15 平安科技(深圳)有限公司 表格识别方法、装置、计算机设备和存储介质
CN111382717A (zh) * 2020-03-17 2020-07-07 腾讯科技(深圳)有限公司 一种表格识别方法、装置和计算机可读存储介质
CN111860257A (zh) * 2020-07-10 2020-10-30 上海交通大学 融合多种文本特征及几何信息的表格识别方法及系统
CN112381010A (zh) * 2020-11-17 2021-02-19 深圳壹账通智能科技有限公司 表格结构的还原方法、系统、计算机设备及存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118351543A (zh) * 2024-06-18 2024-07-16 南昌大学第一附属医院 一种医疗检验单的数据信息提取分析方法

Also Published As

Publication number Publication date
CN112949443A (zh) 2021-06-11
CN112949443B (zh) 2023-07-25

Similar Documents

Publication Publication Date Title
WO2022178994A1 (zh) 表格结构识别方法、装置、电子设备及存储介质
CN111813963B (zh) 知识图谱构建方法、装置、电子设备及存储介质
US10789461B1 (en) Automated systems and methods for textual extraction of relevant data elements from an electronic clinical document
CN104835098A (zh) 一种病历电子数据识别方法及系统
WO2022105172A1 (zh) Pdf文档跨页表格合并方法、装置、电子设备及存储介质
CN111512315A (zh) 文档元数据的按块提取
CN113051356A (zh) 开放关系抽取方法、装置、电子设备及存储介质
WO2021189827A1 (zh) 识别模糊图像的方法、装置、设备及计算机可读存储介质
CN113408323B (zh) 表格信息的提取方法、装置、设备及存储介质
CN113672781A (zh) 数据查询方法、装置、电子设备及存储介质
WO2022100032A1 (zh) 系统分析可视化方法、装置、电子设备及计算机可读存储介质
WO2022227192A1 (zh) 图像分类方法、装置、电子设备及介质
CN113434674A (zh) 数据解析方法、装置、电子设备及可读存储介质
WO2022142106A1 (zh) 文本分析方法、装置、电子设备及可读存储介质
CN112528013A (zh) 文本摘要提取方法、装置、电子设备及存储介质
CN111695330B (zh) 生成表格的方法、装置、电子设备及计算机可读存储介质
WO2023071127A1 (zh) 政策推荐方法、装置、设备及存储介质
CN113360139A (zh) 前端框架的集成方法、装置、电子设备及存储介质
CN113971044A (zh) 组件文档生成方法、装置、设备及可读存储介质
CN113204698A (zh) 新闻主题词生成方法、装置、设备及介质
US20230385559A1 (en) Automated methods and systems for retrieving information from scanned documents
CN113468175A (zh) 数据压缩方法、装置、电子设备及存储介质
CN113505273A (zh) 基于重复数据筛选的数据排序方法、装置、设备及介质
CN112529743A (zh) 合同要素抽取方法、装置、电子设备及介质
CN116578696A (zh) 文本摘要生成方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21927427

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21927427

Country of ref document: EP

Kind code of ref document: A1