CN112381010A - Table structure restoration method, system, computer equipment and storage medium - Google Patents
Table structure restoration method, system, computer equipment and storage medium Download PDFInfo
- Publication number
- CN112381010A CN112381010A CN202011290469.5A CN202011290469A CN112381010A CN 112381010 A CN112381010 A CN 112381010A CN 202011290469 A CN202011290469 A CN 202011290469A CN 112381010 A CN112381010 A CN 112381010A
- Authority
- CN
- China
- Prior art keywords
- node
- training
- recognition model
- identified
- nodes
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
- G06V30/412—Layout analysis of documents structured with printed lines or input boxes, e.g. business forms or tables
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of artificial intelligence, and discloses a method, a system, computer equipment and a computer storage medium for restoring a table structure, wherein the method comprises the following steps: constructing node characteristics of each table node in the table to be identified; calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance; and carrying out post-processing on the node relation to restore the table structure of the table to be identified. In addition, the invention also relates to a block chain technology, and the table identification model can be stored in the block chain. The method can automatically identify the table structure and restore the table structure without the image of the table data having higher quality, removes the dependence of the identified and restored table structure on the image quality, and improves the identification accuracy of the table structure and the table restoring efficiency.
Description
Technical Field
The present invention relates to the field of artificial intelligence technologies, and in particular, to a method, a system, a computer device, and a computer storage medium for restoring a table structure.
Background
The table is a form of structured data that is both simple and standardized. The table data can be quickly understood by a user due to the characteristic of clear structure, so that the digital information such as financial data, statistical data and the like is usually presented in a table form, and particularly, the table data is taken as key information in the financial data and is more and more emphasized by financial personnel in the financial data processing process. While table data extraction is a processing operation that is already common in various fields, extracting data information in a table based on manual operation is usually a tedious and time-consuming process. Therefore, there is a need in various fields for an extraction method capable of automatically completing form data instead of manual operation.
In the existing products, the table structure is mostly restored by an image recognition method, but the method highly depends on detection and recognition of lines in the table, so that when the background of the table image is relatively complex or the printing of the lines in the table is intermittent and fuzzy, the accuracy of the table structure restoration is greatly reduced. Meanwhile, for the table without the frame line, the position of the line needs to be estimated by some other methods, and then the table structure is restored. Thus. In the existing method for restoring the table structure, the restoration effect of the table structure depends on the image quality, and when the image quality is not high, the restoration effect of the table structure is relatively poor.
In summary, it is difficult for the conventional method to accurately restore the table structure without depending on the image quality.
Disclosure of Invention
The invention mainly aims to provide a table structure restoration method, a table structure restoration device, computer equipment and a computer storage medium, aiming at removing the dependence of table structure identification and restoration on images, thereby realizing efficient and accurate restoration of a table structure for information extraction.
In order to achieve the above object, an embodiment of the present invention provides a table structure restoring method, where the table structure restoring method includes:
constructing node characteristics of each table node in the table to be identified;
calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance;
and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
Optionally, the graph convolution network includes, but is not limited to, a two-layer graph convolution network, and before the step of invoking a preset table identification model for the table identification model to output a node relationship between each table node according to the node feature, the method further includes:
and constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model.
Optionally, the training data includes a training sample table and a training test table,
the step of constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model comprises the following steps of:
obtaining public table data and preprocessing the public table data to construct and obtain the training sample table and the training test table;
constructing first node characteristics of each table node in the training sample table, and inputting the first node characteristics into a two-layer graph convolution network for training to obtain a table identification model to be confirmed;
constructing second node characteristics of each table node in the training test table, and inputting the second node characteristics into the table recognition model to be confirmed so as to test the table recognition model to be confirmed and determine a test result;
and if the test result meets the preset model recognition condition, taking the to-be-confirmed form recognition model as a trained form recognition model.
Optionally, after the step of constructing second node features of each table node in the training test table, and inputting the second node features into the to-be-confirmed table recognition model, so as to perform a test on the to-be-confirmed table recognition model and determine a test result, the method further includes:
and if the test result does not accord with the preset model identification condition, constructing a new training sample table to retrain the two-layer graph convolution network.
Optionally, the step of constructing node features of table nodes in the table to be identified includes:
establishing cell characteristics of each table node in the table to be identified;
respectively extracting the cell features of the table nodes adjacent to the table nodes to form a cell feature set;
and carrying out aggregation operation on the cell feature set to obtain respective node features of the table nodes.
Optionally, the table recognition model is stored in a block chain, the node relationships include, but are not limited to, horizontal, vertical and irrelevant,
the step of calling a preset form recognition model to enable the form recognition model to output the node relation among the form nodes according to the node characteristics comprises the following steps:
extracting the table identification model from the blockchain;
and inputting the respective node characteristics of each table node in the table to be recognized into the table recognition model, so that after the table recognition model is trained and calculated based on the node characteristics, the horizontal, vertical or irrelevant node relation between two table nodes of each table node is output.
Optionally, the step of performing post-processing on the node relationship to restore the table structure of the table to be identified includes:
and carrying out post-processing on the node relation by utilizing a preset maximal clustering algorithm so as to restore the table structure of the table to be identified to obtain a complete table.
In addition, to achieve the above object, the present invention further provides a table structure restoring system, including:
the building module is used for building node characteristics of each table node in the table to be identified;
the identification module is used for calling a preset form identification model so that the form identification model can output the node relation among the form nodes according to the node characteristics, wherein the form identification model is obtained in advance based on graph convolution network training;
and the restoring module is used for carrying out post-processing on the node relation so as to restore the table structure of the table to be identified.
Further, to achieve the above object, the present invention also provides a computer apparatus comprising: a memory, a processor, a communication bus, and a recovery program of a table structure stored on the memory,
the communication bus is used for realizing communication connection between the processor and the memory;
the processor is used for executing the table structure restoring program to realize the following steps:
constructing node characteristics of each table node in the table to be identified;
calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance;
and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
Further, to achieve the above object, the present invention also provides a computer storage medium storing one or more programs executable by one or more processors for:
constructing node characteristics of each table node in the table to be identified;
calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance;
and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
The table structure reduction method, the table structure reduction system, the computer equipment and the calculation storage medium provided by the invention have the advantages that the node characteristics of each table node in the table to be identified are constructed; calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance; and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
In the process of needing to restore the table structure of the table data, the table data is used as the table to be recognized and the respective node characteristics of each table node in the table to be recognized are constructed, then a table recognition model obtained in advance based on graph convolution network training is called, the respective node characteristics of each table node are used as input, the table recognition model predicts and outputs the node relation among each table node in the table to be recognized according to the node characteristics, and finally, post-processing operation is carried out on the node relation among each table node so as to restore the table structure of the table to be recognized.
The invention realizes that the table recognition model obtained based on graph convolution network training predicts and outputs the node relation according to the node characteristics of the table to be recognized, and then carries out post-processing operation aiming at the node relation so as to restore and obtain the table structure of the table to be recognized. Therefore, the table structure can be automatically identified and restored without the image of the table data having higher quality, the dependence of the identified and restored table structure on the image quality is eliminated, and the identification accuracy and the table restoration efficiency of the table structure are improved.
Drawings
FIG. 1 is a schematic structural diagram of a hardware operating environment of a computer device according to a method of an embodiment of the present invention;
FIG. 2 is a flowchart illustrating an embodiment of a method for reducing a table structure according to the present invention;
FIG. 3 is a schematic diagram of an application scenario according to an embodiment of the table structure reduction method of the present invention;
FIG. 4 is a functional block diagram of the table structured recovery system of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The main solution of the embodiment of the invention is as follows: constructing node characteristics of each table node in the table to be identified; calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance; and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
Since the table is a form of structured data, it is both simple and standardized. The table data can be quickly understood by a user due to the characteristic of clear structure, so that the digital information such as financial data, statistical data and the like is usually presented in a table form, and particularly, the table data is taken as key information in the financial data and is more and more emphasized by financial personnel in the financial data processing process. While table data extraction is a processing operation that is already common in various fields, extracting data information in a table based on manual operation is usually a tedious and time-consuming process. Therefore, there is a need in various fields for an extraction method capable of automatically completing form data instead of manual operation.
In the existing products, the table structure is mostly restored by an image recognition method, but the method highly depends on detection and recognition of lines in the table, so that when the background of the table image is relatively complex or the printing of the lines in the table is intermittent and fuzzy, the accuracy of the table structure restoration is greatly reduced. Meanwhile, for the table without the frame line, the position of the line needs to be estimated by some other methods, and then the table structure is restored. Thus. In the existing method for restoring the table structure, the restoration effect of the table structure depends on the image quality, and when the image quality is not high, the restoration effect of the table structure is relatively poor.
In summary, it is difficult for the conventional method to accurately restore the table structure without depending on the image quality.
According to the solution provided by the invention, in the process of needing to restore the table structure of the table data, the table data is used as the table to be recognized and the respective node characteristics of each table node in the table to be recognized are constructed, then, a table recognition model obtained in advance based on graph convolution network training is called, the respective node characteristics of each table node are used as input, the table recognition model predicts and outputs the node relation among each table node in the table to be recognized according to the node characteristics, and finally, the post-processing operation is carried out on the node relation among each table node, so that the table structure of the table to be recognized is restored.
The invention realizes that the table recognition model obtained based on graph convolution network training predicts and outputs the node relation according to the node characteristics of the table to be recognized, and then carries out post-processing operation aiming at the node relation so as to restore and obtain the table structure of the table to be recognized. Therefore, the table structure can be automatically identified and restored without the image of the table data having higher quality, the dependence of the identified and restored table structure on the image quality is eliminated, and the identification accuracy and the table restoration efficiency of the table structure are improved.
As shown in fig. 1, fig. 1 is a schematic device structure diagram of a hardware operating environment of a computer device according to an embodiment of the present invention.
The computer equipment of the embodiment of the invention can be terminal equipment such as a PC, a smart phone, a tablet computer, a portable computer and the like.
As shown in fig. 1, the computer apparatus may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a storage device separate from the processor 1001.
Optionally, the computer device may further include a camera, RF (Radio Frequency) circuitry, sensors, audio circuitry, a WiFi module, and the like. Such as light sensors, motion sensors, and other sensors. In particular, the light sensor may include an ambient light sensor that adjusts the brightness of the display screen based on the ambient light level and a proximity sensor that turns off the display screen and/or backlight when the device is moved to the ear. As one of the motion sensors, the gravity acceleration sensor can detect the magnitude of acceleration in each direction (generally, three axes), detect the magnitude and direction of gravity when the device is stationary, and can be used for applications of recognizing the device posture (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer and tapping), and the like; of course, the mobile terminal may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, and an infrared sensor, which are not described herein again.
Those skilled in the art will appreciate that the device architecture shown in fig. 1 is not intended to be limiting as the computer device may include more or less components than shown, or some components may be combined, or a different arrangement of components in other embodiments.
As shown in fig. 1, a memory 1005, which is a kind of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and a table structure restoring program.
In the computer device shown in fig. 1, the network interface 1004 is mainly used for connecting to a backend server and communicating with the backend server; the user interface 1003 is mainly used for connecting a client (user side) and performing data communication with the client; and the processor 1001 may be configured to call a restore procedure of the table structure stored in the memory 1005, and perform the following steps:
constructing node characteristics of each table node in the table to be identified;
calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance;
and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
Further, the graph convolution network includes, but is not limited to, a two-layer graph convolution network, and the processor 1001 may be configured to invoke a restoring program of a table structure stored in the memory 1005, and before executing the step of invoking a preset table identification model for the table identification model to output a node relationship between each table node according to the node feature, further execute the following steps:
and constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model.
Further, the training data includes a training sample table and a training test table, and the processor 1001 may be configured to call a restoring program of a table structure stored in the memory 1005, and further perform the following steps:
obtaining public table data and preprocessing the public table data to construct and obtain the training sample table and the training test table;
constructing first node characteristics of each table node in the training sample table, and inputting the first node characteristics into a two-layer graph convolution network for training to obtain a table identification model to be confirmed;
constructing second node characteristics of each table node in the training test table, and inputting the second node characteristics into the table recognition model to be confirmed so as to test the table recognition model to be confirmed and determine a test result;
and if the test result meets the preset model recognition condition, taking the to-be-confirmed form recognition model as a trained form recognition model.
Further, the processor 1001 may be configured to invoke a restoring program of the table structure stored in the memory 1005, and after the step of constructing the second node feature of each table node in the training test table, and inputting the second node feature into the to-be-confirmed table recognition model, so as to perform a test on the to-be-confirmed table recognition model and determine a test result, perform the following steps:
and if the test result does not accord with the preset model identification condition, constructing a new training sample table to retrain the two-layer graph convolution network.
Further, the processor 1001 may be configured to call a restoring program of the table structure stored in the memory 1005, and further perform the following steps:
establishing cell characteristics of each table node in the table to be identified;
respectively extracting the cell features of the table nodes adjacent to the table nodes to form a cell feature set;
and carrying out aggregation operation on the cell feature set to obtain respective node features of the table nodes.
Further, the table recognition model is stored in a block chain, the node relationships include, but are not limited to, horizontal, vertical and irrelevant, and the processor 1001 may be configured to call a restoring program of the table structure stored in the memory 1005, and further perform the following steps:
extracting the table identification model from the blockchain;
and inputting the respective node characteristics of each table node in the table to be recognized into the table recognition model, so that after the table recognition model is trained and calculated based on the node characteristics, the horizontal, vertical or irrelevant node relation between two table nodes of each table node is output.
Further, the processor 1001 may be configured to call a restoring program of the table structure stored in the memory 1005, and further perform the following steps:
and carrying out post-processing on the node relation by utilizing a preset maximal clustering algorithm so as to restore the table structure of the table to be identified to obtain a complete table.
The specific embodiment of the computer device related to the method for restoring the table structure of the present invention is substantially the same as each specific embodiment of the method for restoring the table structure described below, and is not described herein again.
The invention provides a table structure reduction method.
Referring to fig. 2, fig. 2 is a schematic flow chart of a table structure reduction method according to a first embodiment of the present invention, in this embodiment, the table structure reduction method includes:
s100, constructing node characteristics of each table node in a table to be identified;
in the process of carrying out a table structure on table data, the terminal device takes the table data as a table to be identified and sequentially constructs respective node characteristics of each table node aiming at each table node in the table to be identified.
It should be noted that, in this embodiment, the form to be identified is any form data that needs to be identified and/or restored and is received by the terminal device, for example, the form data of an enterprise financial affairs, the form data of an enterprise personnel statistics, and the like, it should be understood that, based on different design requirements of practical applications, the form to be identified may be different from other types of form data listed herein in other feasible real-time manners, and the method for restoring the form structure of the present invention is not specifically limited to the type of the form data of the form to be identified.
In addition, in this embodiment, each table node in the table to be recognized may be specifically each cell in the table to be recognized, and it should be understood that, based on different design requirements of practical applications, in other feasible real-time manners, other elements different from the cell mentioned here in the table to be recognized may also be used as the table node, and similarly, the method for restoring the table structure of the present invention is not specifically limited to the type of the element of the table node in the table to be recognized.
Further, in a possible embodiment, the step S100 may include:
step S101, establishing cell characteristics of each table node in the table to be identified;
it should be noted that, in this embodiment, the cell features of each table node include, but are not limited to: the cell coordinates, the cell sizes, and the frequencies of the cell text types may specifically be, for example, the cell coordinates may specifically be a horizontal ordinate and a vertical ordinate of an upper left corner of a text box position of each table node (i.e., a cell), a horizontal ordinate and a vertical ordinate of a lower right corner of a text box position of a cell, and a coordinate of a center position of a text box of a cell in the table to be recognized, the cell sizes may specifically be widths and heights of text boxes of each cell in the table to be recognized, the frequencies of the cell text types may specifically be, in the table to be recognized, the frequencies of the text types in each cell are numbers, the frequencies of the text types are letters, the frequencies of the text types are spaces, or the frequencies of the text types are other types of numerical values.
The terminal equipment sequentially identifies and extracts the cell coordinates, the cell sizes and the cell text type frequencies of all the table nodes aiming at all the table nodes in the table to be identified, then integrates the cell coordinates, the cell sizes and the cell text type frequencies of all the table nodes as the cell features of the table nodes, and further establishes and obtains the cell features of all the table nodes in the table to be identified.
Specifically, for example, the terminal device identifies and extracts, by using an existing mature OCR (Optical Character Recognition) technology, each table node (i.e. cell) in the table to be identified by aiming at a picture containing the table to be identified: n _ u _1, N _ u _2, N _ u _3.. N _ u _ i (i is a positive integer), the abscissa and ordinate of the upper left corner of the text box position of each cell: (x1, y1), the abscissa of the lower right corner of the text box position: (x2, y2), coordinates of the center position of the text box: (center _ x, center _ y), width and height of the text box: (w, h), and the frequency with which the text type is numeric in the cell: num, frequency with text type in cell as letter: char, frequency with text type in cell as space: frequency of space or other types of values for text in cells: other, the terminal device then integrates the (x1, y1), (x2, y2), (center _ x, center _ y), (w, h) and num, char, space, other data belonging to the same table node (i.e., cell) as a cell feature of each table node (i.e., cell): f (N _ u _ i) [ x1, y1, x2, y2, w, h, center _ x, center _ y, num, char, space, other ].
Step S102, respectively extracting the cell features of the table nodes adjacent to the table nodes to form a cell feature set;
after the respective cell features of all the table nodes in the table to be identified are established and obtained, the terminal equipment respectively extracts the respective cell features of the table nodes adjacent to the table nodes in sequence aiming at each table node, and combines the cell features into a cell feature set.
Specifically, for example, after the terminal device obtains the respective cell feature of each table node (i.e., cell) in the table to be identified through integration and establishment, the terminal device sequentially targets, by using an existing mature KNN (K-nearest neighbor algorithm) technique, other several neighboring table nodes (neighboring cells, usually 3, 6, or 9) near (closest to) each table node (i.e., cell) N _ u _ i, and then the terminal device takes the respective cell features f (N _ u _1), f (N _ u _2), …, f (N _ u _9) of the 3, 6, or 9 neighboring table nodes (neighboring cells) near (i.e., cell) N _ u _ i as a cell feature set.
Step S103, carrying out aggregation operation on the cell feature set to obtain respective node features of the table nodes.
And the terminal equipment carries out aggregation operation on the cell feature set corresponding to each table node respectively, and takes the result of the aggregation operation as the node feature of the table node, so that the respective node features of all the table nodes in the table to be identified are constructed and obtained.
Specifically, for example, the terminal device performs an aggregation operation on the cell feature set f (N _ u _1), f (N _ u _2), …, and f (N _ u _9) corresponding to each table node (i.e., cell) N _ u _ i in the table to be identified, so as to obtain the node feature f (N) of the table node (i.e., cell) N _ u _ i. The specific operation process is as follows:
f(N)＝Aggregate(f(N_u_1)，f(N_u_2)，…，f(N_u_9)，f(N))；
N_u＝Neighbor(N)。
it should be noted that, in this embodiment, Aggregate is a well-established aggregation operation, and Aggregate generally refers to: summing or averaging.
Step S200, calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained in advance based on graph convolution network training;
after the terminal device constructs and obtains the respective node characteristics of each table node in the table to be recognized, the terminal device calls a table recognition model which is obtained in advance based on graph convolution network training, and the respective node characteristics of each table node are input into the table recognition model, so that the table recognition model carries out training calculation based on the node characteristics, the node relation among the table nodes in the table to be recognized is predicted, and the predicted node relation is output.
It should be noted that, in this embodiment, the table identification model invoked by the terminal device may be a table identification model obtained by the terminal device based on graph convolution network training in advance, and then the table identification model is stored in a certain node of a block chain created in advance for subsequent invocation.
Further, in a possible embodiment, the step S200 may include:
step S201, extracting the table recognition model from the block chain;
step S202, inputting the node characteristics of each table node in the table to be recognized into the table recognition model, so that after the table recognition model performs training calculation based on the node characteristics, the horizontal, vertical or irrelevant node relation between two table nodes of each table node is output.
It should be noted that, in this embodiment, the table identification model may specifically be based on a two-layer graph convolution network stacked with two identical (GCNLNReLUGCNLN) structures in advance: GCNLNReLUGCNLN training.
Specifically, for example, referring to the application scenario shown in fig. 3, after the terminal device constructs the node feature f (N) of each table node (i.e. cell) N _ u _ i in the table to be recognized, the terminal device extracts the table recognition model from the block chain nodes pre-stored with the table recognition model trained based on the graph-rolling network, and then the terminal device inputs the node feature f (N) of each constructed table node (i.e. cell) N _ u _ i into the extracted table recognition model stacked with two identical (gcnleulgcn) structures, so that a residual error connection is formed by directly adding the output feature (f) of the first node feature (f) through the first (gcnleulgcn) structure stacked with the input feature f (N) through the second (gcnleulgcn) structure stacked with the node feature f (N) through the first (gcnleulgcn) structure, therefore, the node characteristics f (N) of any two table nodes (namely the cells) N _ u _ i are merged to predict the horizontal, vertical or irrelevant node relation of the edge between the two table nodes (namely the cells) N _ u _ i, and the label "0" corresponding to the "horizontal" node relation, the label "1" corresponding to the "vertical" node relation or the label "2" corresponding to the "irrelevant" node relation are output.
In this embodiment, by using the node of the block chain of the table recognition model obtained by training based on the graph convolution network in advance, not only the stability of the table recognition model can be ensured, but also the response aggressiveness and accuracy of the terminal device in extracting the table recognition model can be ensured, and the efficiency of identifying and restoring the table structure of the table to be recognized by training the table recognition model based on the graph convolution network can be further improved.
Step S300, carrying out post-processing aiming at the node relation so as to restore the table structure of the table to be identified.
After the terminal device trains, predicts and outputs the node relationship between each table node in the table to be recognized according to the node characteristics of each table node in the table to be recognized based on calling a table recognition model obtained by pre-training, the terminal device further performs post-processing on the node relationship so as to form a complete table based on the node relationship between two table nodes in the table to be recognized.
Further, in a possible embodiment, the step S300 may include:
step S301, post-processing is carried out on the node relation by utilizing a preset maximal clustering algorithm so as to restore the table structure of the table to be identified to obtain a complete table.
It should be noted that, in this embodiment, the maximal clique algorithm may specifically be a mature maximal clique algorithm in a graph theory, and it should be understood that, based on different design requirements of practical applications, in other feasible real-time manners, a terminal device may also use other maximal clique algorithms different from those in this embodiment to perform post-processing operations on node relationships between table nodes in a table to be identified.
Specifically, for example, after the terminal device predicts the horizontal, vertical or irrelevant node relationship of the edge between two table nodes (i.e., cells) N _ u _ i according to the node features f (N) of any two table nodes (i.e., cells) N _ u _ i in the table to be recognized by merging based on the table recognition model, and outputs the label "0" corresponding to the "horizontal" node relationship, the label "1" corresponding to the "vertical" node relationship, or the label "2" corresponding to the "irrelevant" node relationship, so as to determine the node relationship between all table nodes (i.e., cells) N _ u _ i in the table to be recognized, the terminal device further calculates the row extremely large cliques and column extremely large cliques between all table nodes (i.e., cells) N _ u _ i according to the node relationship between all table nodes (i.e., cells) N _ u _ i in the table to be recognized by using the mature maximal clique algorithm in graph theory, and sorting the maximum cliques according to the position information of each group of maximum cliques in the table, thereby restoring the original logic structure of the table.
In this embodiment, in the process of performing a table structure on table data by using a terminal device, the table data is used as a table to be identified, and respective node features of each table node are sequentially constructed for each table node in the table to be identified; after the terminal equipment constructs and obtains the respective node characteristics of each table node in the table to be recognized, the terminal equipment calls a table recognition model which is obtained in advance based on graph convolution network training, and inputs the respective node characteristics of each table node into the table recognition model, so that the table recognition model carries out training calculation based on the node characteristics, the node relation among the table nodes in the table to be recognized is predicted, and the predicted node relation is output; after the terminal device trains, predicts and outputs the node relationship between each table node in the table to be recognized according to the node characteristics of each table node in the table to be recognized based on calling a table recognition model obtained by pre-training, the terminal device further performs post-processing on the node relationship so as to form a complete table based on the node relationship between two table nodes in the table to be recognized.
The invention realizes that in the process of needing to restore the table structure of the table data, the table recognition model obtained based on graph convolution network training predicts and outputs the node relation according to the node characteristics of the table to be recognized, and then carries out post-processing operation aiming at the node relation so as to restore and obtain the table structure of the table to be recognized. Therefore, the table structure can be automatically identified and restored without the image of the table data having higher quality, the dependence of the identified and restored table structure on the image quality is eliminated, and the identification accuracy and the table restoration efficiency of the table structure are improved.
Further, based on the first embodiment of the table structure restoring method of the present invention, a second embodiment of the table structure restoring method of the present invention is provided, and in the second embodiment of the table structure restoring method of the present invention, before the step S200 calls a preset table recognition model, so that the table recognition model outputs a node relationship between each table node according to the node feature, the table structure restoring method of the present invention may further include:
and S400, constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model.
The terminal device builds training data used for training the graph convolution network in advance, and then trains the two-layer graph convolution network by using the training data, so that the two-layer graph convolution network is trained into a form recognition model used for predicting and recognizing the node relation between two form nodes in a form to be recognized.
Further, in this embodiment, the training data for training the graph convolution network includes, but is not limited to: training the sample table and training the test table, in a possible embodiment, step S400 may include:
step S401, obtaining public form data and preprocessing the public form data to construct and obtain the training sample form and the training test form;
the terminal equipment acquires normally shared public form data from any big data platform, and then preprocesses the public form data, so as to construct and obtain a training sample table and a training test table for training the graph convolution network.
It should be noted that, in this embodiment, the public table data may specifically be a public data set SciTSR published by sharing on any large data platform in the world, and the public data set includes any type of table data.
Specifically, for example, the terminal device obtains the published public data set SciTSR shared by the large data platforms from any large data platform, and then, the terminal device extracts 15000 arbitrary types of table data (i.e., 15000 tables) from the public data set SciTSR, and uses 12000 table data of the 15000 table data as training sample tables, and another 3000 table data of the 15000 table data as training test tables.
Step S402, constructing first node characteristics of each table node in the training sample table, and inputting the first node characteristics into a two-layer graph convolution network for training to obtain a table identification model to be confirmed;
the terminal equipment constructs first node characteristics of all table nodes in the table aiming at each table in the constructed training sample tables, constructs labels of node relations among all table nodes, inputs the first node characteristics of all table nodes in each table and the labels of the node relations among all table nodes into a two-layer graph convolution network together for training until the training is finished, and finally, the terminal equipment takes the trained two-layer graph convolution network as a table identification model to be confirmed.
It should be noted that, in this embodiment, a process of the terminal device constructing, for each table in the constructed training sample tables, first node features of all table nodes in the table is the same as a process of the terminal device constructing node features of each table node in the table to be identified, and therefore, the first node features of all table nodes in the table are not constructed for each table in the constructed training sample tables for the terminal device again.
Specifically, for example, after constructing and obtaining first node features of all respective table nodes of 12000 training sample tables, the terminal device defines that a label corresponding to a "horizontal" node relationship between two table nodes in each table is "0", "a label corresponding to a" vertical "node relationship between two table nodes in each table is" 1 ", and a label corresponding to an" irrelevant "node relationship is" 2 "; then, the terminal device inputs the first node features of all table nodes of each of the 12000 training sample tables constructed, together with the labels "0", "1", and "2", to a two-layer graph convolution network stacked with two identical (gcnlrelugcnln) structures: in GCNLNReLUGCNLN, the network is thus convolved by the two-layer graph: the GCNLNReLUGCNLN directly adds the output characteristics of the node characteristics calculated by the first (GCNLNReLUGCNLN) structure with the input characteristics of the second (GCNLNReLUGCNLN) structure to form a residual connection, combines the characteristics of any two cells to predict the relationship (horizontal/vertical/no relationship) of the edges between the two cells and outputs corresponding labels '0', '1' and '2', and finally, the two-layer graph convolution network: and the GCNLNReLUGCNLN updates the cell characteristics for training by continuously minimizing a loss function between the prediction relation and the actual annotation relation, so that a to-be-confirmed table identification model is obtained.
Step S403, constructing second node characteristics of each table node in the training test table, and inputting the second node characteristics into the table recognition model to be confirmed so as to test the table recognition model to be confirmed and determine a test result;
the terminal equipment also constructs second node characteristics of all table nodes in the training test table for each table in the constructed training test table, then inputs the second node characteristics of all table nodes in each table into a to-be-confirmed table recognition model obtained after training for a two-layer graph convolution network, so that the to-be-confirmed table recognition model performs training calculation based on the first node characteristics, predicts node relationships among all table nodes in the training test table, outputs a label for identifying the node relationship, and finally compares the label output by the to-be-confirmed table recognition model with real node relationships among all table nodes in the training test table to determine a test result.
It should be noted that, in this embodiment, a process of the terminal device constructing second node features of all table nodes in the table for each table in the constructed training test table is the same as a process of the terminal device constructing node features of each table node in the table to be identified, and therefore, the second node features of all table nodes in the table are not constructed for each table in the constructed training test table for the terminal device again.
And S404, if the test result meets the preset model recognition condition, taking the to-be-confirmed form recognition model as a trained form recognition model.
It should be noted that, in this embodiment, the test result may specifically be that the terminal device compares the label output by the to-be-confirmed table identification model with the real node relationship between each table node in the training test table, so as to obtain the accuracy of predicting the node relationship between all table nodes of the training test table by the to-be-confirmed table identification model. In addition, the model identification condition may specifically be that the identification table identification model preset by the terminal device accurately predicts the lowest prediction accuracy of the node relationship between two table nodes in the table to be identified.
When the terminal device determines that the table recognition model to be confirmed aims at the training test table, and the test result of predicting the node relation between each table node meets the model recognition condition (the accuracy of predicting the node relation between all table nodes of the training test table by the table recognition model to be confirmed is greater than or equal to the minimum prediction accuracy of predicting the node relation between two table nodes in the table to be recognized by accurately predicting the identification table recognition model preset by the terminal device), the terminal device can use the table recognition model to be confirmed as the table recognition model for predicting and recognizing the node relation between two table nodes in the table to be recognized.
Further, in another possible embodiment, in step S403, after constructing second node features of each table node in the training test table, and inputting the second node features into the to-be-confirmed table recognition model, so as to test the to-be-confirmed table recognition model and determine a test result, the method for restoring a table structure of the present invention may further include:
step S405, if the test result does not accord with the preset model recognition condition, a new training sample table is constructed to retrain the two-layer graph convolution network.
When the terminal equipment determines that a test result of predicting the node relation between table nodes in a table recognition model to be confirmed aiming at a training test table does not accord with a model recognition condition (the accuracy of predicting the node relation between all table nodes of the training test table by the table recognition model to be confirmed is smaller than the lowest prediction accuracy of predicting the node relation between two table nodes in the table to be recognized by an identification table recognition model preset by the terminal equipment), the terminal equipment acquires the public table data again to construct a new training sample table and a new training test table, constructs the first node characteristics of each table node in the new training sample table so as to train a two-layer graph convolution network again to obtain a new table recognition model to be confirmed, and then constructs the second node characteristics of each table node in the new training test table, and testing the new table identification model to be confirmed, and circulating the steps until the test result of testing the table identification model to be confirmed meets the model identification condition.
It should be noted that, in this embodiment, the process of the terminal device re-training the two-layer graph convolution network is substantially the same as the process of constructing the training data and training the two-layer graph convolution network by using the training data, and therefore, the process of re-training the two-layer graph convolution network by using the terminal device is not described here again.
In this embodiment, the terminal device builds training data for training the graph convolution network in advance, and then trains the two-layer graph convolution network by using the training data, so that the two-layer graph convolution network is trained into a table recognition model for performing predictive recognition on a node relationship between two table nodes in a table to be recognized. Therefore, the table identification model obtained based on graph convolution network training predicts and outputs the node relation according to the node characteristics of the table to be identified, and then carries out post-processing operation on the node relation so as to restore and obtain the table structure of the table to be identified. Therefore, the table structure can be automatically identified and restored without the image of the table data having higher quality, the dependence of the identified and restored table structure on the image quality is eliminated, and the identification accuracy and the table restoration efficiency of the table structure are improved.
In addition, the present invention further provides a table structure restoring system, please refer to fig. 4, where fig. 4 is a functional module schematic diagram of the table structure restoring system of the present invention, and the table structure restoring system includes:
the building module 101 is configured to build node features of each table node in the table to be identified;
the identification module 102 is configured to call a preset form identification model, so that the form identification model outputs a node relationship between each form node according to the node feature, where the form identification model is obtained in advance based on graph convolution network training;
a restoring module 103, configured to perform post-processing on the node relationship to restore the table structure of the table to be identified.
Optionally, the graph convolution network includes, but is not limited to, a two-layer graph convolution network, and the table structure restoration system of the present invention further includes:
and the model training module is used for constructing training data and training the two-layer graph convolution network by utilizing the training data to obtain a preset form recognition model.
Optionally, the training data includes a training sample table and a training test table, and the model training module includes:
the acquisition unit is used for acquiring public form data and preprocessing the public form data to construct and obtain the training sample form and the training test form;
the first construction unit is used for constructing first node characteristics of each table node in the training sample table and inputting the first node characteristics into a two-layer graph convolution network for training to obtain a table identification model to be confirmed;
the second construction unit is used for constructing second node characteristics of each table node in the training test table, inputting the second node characteristics into the to-be-confirmed table recognition model, and testing the to-be-confirmed table recognition model and determining a test result;
and the determining unit is used for taking the to-be-confirmed form recognition model as a trained form recognition model if the test result meets a preset model recognition condition.
Optionally, the model training module further includes:
and the repeated training unit is used for constructing a new training sample table to retrain the two-layer graph convolution network if the test result does not accord with the preset model identification condition.
Optionally, the building module 101 includes:
the establishing unit is used for establishing the cell characteristics of each table node in the table to be identified;
the first extraction unit is used for respectively extracting the cell features of the table nodes adjacent to the table nodes to form a cell feature set;
and the aggregation unit is used for carrying out aggregation operation on the cell feature set to obtain respective node features of the table nodes.
Optionally, the table identification model is stored in a block chain, the node relationships include, but are not limited to, horizontal, vertical, and irrelevant, and the identification module includes:
a second extraction unit, configured to extract the table identification model from the block chain;
and the structure identification unit is used for inputting the node characteristics of each table node in the table to be identified into the table identification model, so that after the table identification model is trained and calculated based on the node characteristics, the horizontal, vertical or irrelevant node relation between two table nodes of each table node is output.
Optionally, the restoring module 103 is further configured to perform post-processing on the node relationship by using a preset maximal clique algorithm to restore the table structure of the table to be identified to obtain a complete table.
The specific implementation of the table structure reduction system of the present invention is substantially the same as the embodiments of the table structure reduction method, and is not described herein again.
Furthermore, the present invention also provides a computer storage medium storing one or more programs, the one or more programs further executable by one or more processors for:
constructing node characteristics of each table node in the table to be identified;
calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance;
and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
In addition, the one or more programs may be further executed by one or more processors to call a preset form recognition model, so that before the form recognition model outputs the node relationship between the form nodes according to the node characteristics, the one or more programs are further configured to:
and constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model.
Further, the training data includes a training sample table and a training test table, the one or more programs further executable by the one or more processors to:
obtaining public table data and preprocessing the public table data to construct and obtain the training sample table and the training test table;
constructing first node characteristics of each table node in the training sample table, and inputting the first node characteristics into a two-layer graph convolution network for training to obtain a table identification model to be confirmed;
constructing second node characteristics of each table node in the training test table, and inputting the second node characteristics into the table recognition model to be confirmed so as to test the table recognition model to be confirmed and determine a test result;
and if the test result meets the preset model recognition condition, taking the to-be-confirmed form recognition model as a trained form recognition model.
In addition, the one or more programs may be further executed by the one or more processors to construct second node features of each table node in the training test table, and input the second node features into the to-be-confirmed table recognition model, so as to perform a test on the to-be-confirmed table recognition model and determine a test result, and further configured to:
and if the test result does not accord with the preset model identification condition, constructing a new training sample table to retrain the two-layer graph convolution network.
Further, the table identification model is stored in a blockchain, the node relationships include, but are not limited to, horizontal, vertical, and irrelevant, the one or more programs further executable by the one or more processors for:
extracting the table identification model from the blockchain;
and inputting the respective node characteristics of each table node in the table to be recognized into the table recognition model, so that after the table recognition model is trained and calculated based on the node characteristics, the horizontal, vertical or irrelevant node relation between two table nodes of each table node is output.
The specific implementation of the computer storage medium of the present invention is substantially the same as the embodiments of the table structure restoring method described above, and is not described herein again.
It should be noted that the blockchain in the present invention is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like. Further, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only an alternative embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A method for restoring a table structure is characterized by comprising the following steps:
constructing node characteristics of each table node in the table to be identified;
calling a preset form recognition model to enable the form recognition model to output node relations among the form nodes according to the node characteristics, wherein the form recognition model is obtained by training based on a graph convolution network in advance;
and carrying out post-processing on the node relation to restore the table structure of the table to be identified.
2. A table structure restoration method as claimed in claim 1, wherein the graph convolution network includes but is not limited to a two-layer graph convolution network, and before the step of calling a preset table identification model for the table identification model to output the node relationship between the table nodes according to the node characteristics, the method further includes:
and constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model.
3. The form structure restoration method according to claim 2, wherein the training data includes a training sample form and a training test form,
the step of constructing training data and training the two-layer graph convolution network by using the training data to obtain a preset form recognition model comprises the following steps of:
obtaining public table data and preprocessing the public table data to construct and obtain the training sample table and the training test table;
constructing first node characteristics of each table node in the training sample table, and inputting the first node characteristics into a two-layer graph convolution network for training to obtain a table identification model to be confirmed;
constructing second node characteristics of each table node in the training test table, and inputting the second node characteristics into the table recognition model to be confirmed so as to test the table recognition model to be confirmed and determine a test result;
and if the test result meets the preset model recognition condition, taking the to-be-confirmed form recognition model as a trained form recognition model.
4. The method for restoring table structure according to claim 3, wherein after the step of constructing the second node feature of each table node in the training test table, and inputting the second node feature into the table recognition model to be confirmed, so as to perform a test on the table recognition model to be confirmed and determine a test result, the method further comprises:
and if the test result does not accord with the preset model identification condition, constructing a new training sample table to retrain the two-layer graph convolution network.
5. The method for restoring table structure according to claim 1, wherein the step of constructing the node feature of each table node in the table to be identified includes:
establishing cell characteristics of each table node in the table to be identified;
respectively extracting the cell features of the table nodes adjacent to the table nodes to form a cell feature set;
and carrying out aggregation operation on the cell feature set to obtain respective node features of the table nodes.
6. The method for tabular structure reduction of claim 1, wherein the tabular recognition model is stored in a blockchain, the nodal relationships include but are not limited to horizontal, vertical, and uncorrelated,
the step of calling a preset form recognition model to enable the form recognition model to output the node relation among the form nodes according to the node characteristics comprises the following steps:
extracting the table identification model from the blockchain;
and inputting the respective node characteristics of each table node in the table to be recognized into the table recognition model, so that after the table recognition model is trained and calculated based on the node characteristics, the horizontal, vertical or irrelevant node relation between two table nodes of each table node is output.
7. The table structure restoring method according to claim 1, wherein the step of performing post-processing on the node relationship to restore the table structure of the table to be identified includes:
and carrying out post-processing on the node relation by utilizing a preset maximal clustering algorithm so as to restore the table structure of the table to be identified to obtain a complete table.
8. A system for restoring a table structure, the system comprising:
the building module is used for building node characteristics of each table node in the table to be identified;
the identification module is used for calling a preset form identification model so that the form identification model can output the node relation among the form nodes according to the node characteristics, wherein the form identification model is obtained in advance based on graph convolution network training;
and the restoring module is used for carrying out post-processing on the node relation so as to restore the table structure of the table to be identified.
9. A computer device, characterized in that the computer device comprises: a memory, a processor, a communication bus, and a recovery program of a table structure stored on the memory,
the communication bus is used for realizing communication connection between the processor and the memory;
the processor is configured to execute the internet-based table structure restoring program to implement the steps of the table structure restoring method according to any one of claims 1 to 7.
10. A computer storage medium, characterized in that the computer storage medium has stored thereon a table structure restoring program, which when executed by a processor implements the steps of the table structure restoring method according to any one of claims 1 to 7.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011290469.5A CN112381010A (en) | 2020-11-17 | 2020-11-17 | Table structure restoration method, system, computer equipment and storage medium |
| PCT/CN2021/109491 WO2022105297A1 (en) | 2020-11-17 | 2021-07-30 | Table structure recovery method and system, computer device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011290469.5A CN112381010A (en) | 2020-11-17 | 2020-11-17 | Table structure restoration method, system, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112381010A true CN112381010A (en) | 2021-02-19 |
Family
ID=74585025
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011290469.5A Pending CN112381010A (en) | 2020-11-17 | 2020-11-17 | Table structure restoration method, system, computer equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112381010A (en) |
| WO (1) | WO2022105297A1 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112949443A (en) * | 2021-02-24 | 2021-06-11 | 平安科技（深圳）有限公司 | Table structure identification method and device, electronic equipment and storage medium |
| CN114495140A (en) * | 2022-04-14 | 2022-05-13 | 安徽数智建造研究院有限公司 | Method, system, device, medium and program product for extracting information of table |
| WO2022105297A1 (en) * | 2020-11-17 | 2022-05-27 | 深圳壹账通智能科技有限公司 | Table structure recovery method and system, computer device, and storage medium |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116824611B (en) * | 2023-08-28 | 2024-04-05 | 星汉智能科技股份有限公司 | Table structure identification method, electronic device, and computer-readable storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110532834B (en) * | 2018-05-24 | 2022-12-23 | 北京庖丁科技有限公司 | Table extraction method, device, equipment and medium based on rich text format document |
| CN109344823B (en) * | 2018-09-11 | 2022-06-07 | 福建天晴在线互动科技有限公司 | OCR deep learning method based on block chain mechanism and storage medium |
| CN111640130A (en) * | 2020-05-29 | 2020-09-08 | 深圳壹账通智能科技有限公司 | Table reduction method and device |
| CN111860257B (en) * | 2020-07-10 | 2022-11-11 | 上海交通大学 | Table identification method and system fusing multiple text features and geometric information |
| CN112381010A (en) * | 2020-11-17 | 2021-02-19 | 深圳壹账通智能科技有限公司 | Table structure restoration method, system, computer equipment and storage medium |
-
2020
- 2020-11-17 CN CN202011290469.5A patent/CN112381010A/en active Pending
-
2021
- 2021-07-30 WO PCT/CN2021/109491 patent/WO2022105297A1/en active Application Filing
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022105297A1 (en) * | 2020-11-17 | 2022-05-27 | 深圳壹账通智能科技有限公司 | Table structure recovery method and system, computer device, and storage medium |
| CN112949443A (en) * | 2021-02-24 | 2021-06-11 | 平安科技（深圳）有限公司 | Table structure identification method and device, electronic equipment and storage medium |
| WO2022178994A1 (en) * | 2021-02-24 | 2022-09-01 | 平安科技（深圳）有限公司 | Table structure recognition method and apparatus, electronic device, and storage medium |
| CN114495140A (en) * | 2022-04-14 | 2022-05-13 | 安徽数智建造研究院有限公司 | Method, system, device, medium and program product for extracting information of table |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022105297A1 (en) | 2022-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112381010A (en) | Table structure restoration method, system, computer equipment and storage medium | |
| WO2022037541A1 (en) | Image processing model training method and apparatus, device, and storage medium | |
| CN108334805B (en) | Method and device for detecting document reading sequence | |
| CN103403739A (en) | Method and system for comparing images | |
| CN112001373B (en) | Article identification method and device and storage medium | |
| CN114402369A (en) | Human body posture recognition method and device, storage medium and electronic equipment | |
| CN111368636A (en) | Object classification method and device, computer equipment and storage medium | |
| CN111881740B (en) | Face recognition method, device, electronic equipment and medium | |
| CN111144215A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| EP4343616A1 (en) | Image classification method, model training method, device, storage medium, and computer program | |
| CN114862845A (en) | Defect detection method, device and equipment for mobile phone touch screen and storage medium | |
| CN110807110A (en) | Image searching method and device combining local and global features and electronic equipment | |
| CN111222585A (en) | Data processing method, device, equipment and medium | |
| CN111783777B (en) | Image processing method, apparatus, electronic device, and computer readable medium | |
| CN110392115B (en) | Block chain node management method, device and readable storage medium | |
| CN110245669B (en) | Palm key point identification method, device, terminal and readable storage medium | |
| CN115439700B (en) | Image processing method and device and machine-readable storage medium | |
| CN111476144A (en) | Pedestrian attribute identification model determination method and device and computer readable storage medium | |
| CN114187488B (en) | Image processing method, device, equipment and medium | |
| CN113344994B (en) | Image registration method, device, electronic equipment and storage medium | |
| CN115170838A (en) | Data screening method and device | |
| CN114022658A (en) | Target detection method, device, storage medium and terminal | |
| CN114385714A (en) | Feedback-based multi-row-order fusion method, device and equipment and readable storage medium | |
| CN113515701A (en) | Information recommendation method and device | |
| CN113868460A (en) | Image retrieval method, device and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40045376 Country of ref document: HK |
|
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |