Physical examination report form identification method, device, equipment and medium

Abstract

The present application relates to artificial intelligence technology, image recognition, character recognition, table recognition and intelligent medical technology field, disclose a kind of physical examination report table recognition method, device, equipment and medium.The present application includes to the second correction picture of the inclination angle of the physical examination report table picture to be identified is adjusted, obtains;If the second correction picture is the picture of system screenshot type, the table recognition data is determined by carrying out table recognition to the second correction picture through deep learning model;If the second correction picture is the picture of camera shooting type, the table recognition data is determined by carrying out the table recognition to the second correction picture through text block space restoration strategy.The technical scheme of the present application improves the accuracy of physical examination report table recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of artificial intelligence technology, image recognition, character recognition, table recognition and intelligent medical technology, and in particular to a physical examination report table recognition method, device, equipment and medium. BACKGROUND

[0002] Image table restoration is a derivative application of optical character recognition (OCR) for image recognition in the current market. Structured restoration of historical image data is the basis of digital services. For example, in the medical field, patient physical examination reports are often provided in the form of electronic pictures taken or screenshots, or paper physical examination reports. Automatic table restoration can quickly help business personnel enter data in the pictures or paper reports.

[0003] There are many table restoration technologies in the industry at present. The main technical solutions are rule-based methods and statistical-based methods. However, for the single scene of physical examination report data, more statistical-based methods are used, such as deep learning based on labeled data for table restoration. However, such a solution often results in missing recognition items, incorrect recognition content, and misaligned rows and columns in the presence of wrinkles, tilting and bending of paper reports, and the overall recognition accuracy cannot meet business requirements. SUMMARY

[0004] The present application provides a physical examination report table recognition method, device, equipment and medium to solve the technical problem of low accuracy of physical examination report table recognition.

[0005] In a first aspect, the present application provides a physical examination report table recognition method, comprising:

[0006] Obtaining a physical examination report table picture to be recognized;

[0007] Comparing the picture text of the physical examination report table picture to be recognized to determine a large-angle tilt determination result;

[0008] Rotating the physical examination report table picture to be recognized by a specific angle according to the large-angle tilt determination result, so that the tilt angle of the physical examination report table picture to be recognized is less than 90 degrees, to obtain a first corrected picture, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees;

[0009] Determining a tilt angle determination result by performing a tilt angle judgment on the first corrected picture;

[0010] if the inclination angle judgment result is greater than a preset angle, then the first corrected picture is adjusted in rotation to determine a second corrected picture; if the inclination angle judgment result is less than or equal to the preset angle, then the first corrected picture is taken as the second corrected picture;

[0011] if the second corrected picture is a picture of a system screenshot type, then a deep learning model is used to perform table identification on the second corrected picture to determine table identification data;

[0012] if the second corrected picture is a picture of a camera shooting type, then a text block space restoration strategy is used to perform the table identification on the second corrected picture to determine the table identification data.

[0013] In a second aspect, the present application provides a physical examination report table identification device, comprising:

[0014] an acquisition module configured to acquire a physical examination report table picture to be identified;

[0015] a correction module configured to perform picture text comparison on the physical examination report table picture to be identified to determine a large-angle inclination judgment result;

[0016] and further configured to rotate the physical examination report table picture to be identified by a specific angle according to the large-angle inclination judgment result, so that the inclination angle of the physical examination report table picture to be identified is less than 90 degrees, to obtain a first corrected picture, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees;

[0017] and further configured to perform inclination angle judgment on the first corrected picture to determine an inclination angle judgment result;

[0018] and further configured to, if the inclination angle judgment result is greater than a preset angle, then adjust the first corrected picture in rotation to determine a second corrected picture; if the inclination angle judgment result is less than or equal to the preset angle, then take the first corrected picture as the second corrected picture;

[0019] a processing module configured to, if the second corrected picture is a picture of a system screenshot type, then use a deep learning model to perform table identification on the second corrected picture to determine table identification data;

[0020] and further configured to, if the second corrected picture is a picture of a camera shooting type, then use a text block space restoration strategy to perform the table identification on the second corrected picture to determine the table identification data.

[0021] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the steps of the physical examination report form recognition method when executing the computer program.

[0022] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program implements the steps of the physical examination report form recognition method when executed by a processor.

[0023] The scheme implemented by the physical examination report form identification method, device, equipment and medium provided in the application comprises the following steps: obtaining a physical examination report form picture to be identified; performing picture text comparison on the physical examination report form picture to be identified to determine a large-angle tilt determination result; rotating the physical examination report form picture to be identified by a specific angle according to the large-angle tilt determination result, so that the tilt angle of the physical examination report form picture to be identified is less than 90 degrees, and a first corrected picture is obtained, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees. By determining the large-angle tilt determination result, the physical examination report form picture can be preliminarily adjusted in angle, so that the physical examination report form picture is prevented from being completely reversed or having a large angle deviation, and the tilt angle of the physical examination report form picture is less than 90 degrees. The tilt angle of the first corrected picture is determined, and a tilt angle determination result is determined. If the tilt angle determination result is greater than a preset angle, the first corrected picture is rotated and adjusted to determine a second corrected picture. If the tilt angle determination result is less than or equal to the preset angle, the first corrected picture is taken as the second corrected picture. By adjusting the first corrected picture in a small angle, the tilt angle can be less than the preset angle, and the accuracy of subsequent table image identification is ensured. If the second corrected picture is a system screenshot type picture, a table identification data is determined by performing table identification on the second corrected picture by using a deep learning model. If the second corrected picture is a camera shooting type picture, the table identification data is determined by performing the table identification on the second corrected picture by using a text block space restoration strategy. Since the system screenshot type picture does not have paper creases, bending and other problems, the table identification on the second corrected picture by using the deep learning model can improve the identification efficiency while ensuring the identification effect. The camera shooting picture is prone to identification errors due to light, angle, paper creases and other problems, and therefore, by using the text block space restoration strategy, the accuracy of the physical examination report form identification can be improved in the case of paper creases, angle distortion or light shadow. Based on this, the physical examination report form identification method, device, equipment and medium provided in the application adjust the tilt angle of the physical examination report form picture, and different identification strategies are used according to the differences between the system screenshot type picture and the camera shooting type picture, so that the identification efficiency and accuracy are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is an application environment schematic diagram of the physical examination report form identification method in an embodiment of the application;

[0025] Figure 2 is a flowchart of the physical examination report form identification method in an embodiment of the application;

[0026] Figure 3 is a structural schematic diagram of a physical examination report form recognition device in an embodiment of the present application;

[0027] Figure 4 is a structural schematic diagram of a computer device in an embodiment of the present application;

[0028] Figure 5 is another structural schematic diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0030] It should be noted that the terms "first", "second", and the like in the description of the present application and the claims and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or a sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0031] In the description of the present application, the description of the terms "embodiment", "some embodiments", and "optional embodiment" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or implementation are included in at least one embodiment or implementation of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or implementation. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or implementations in a suitable manner.

[0032] The physical examination report form recognition method provided by the embodiments of the present application can be applied to, for example, Figure 1In an application environment of the medical examination report form recognition method, the medical examination report form recognition method can be applied in a client device or a server device, wherein the client communicates with the server through a network. The client can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers. In some embodiments, the client can directly obtain the to-be-recognized medical examination report form picture or obtain the to-be-recognized medical examination report form picture through the server. The server can directly obtain the to-be-recognized medical examination report form picture or obtain the to-be-recognized medical examination report form picture through the client. The medical examination report form picture can be applied in various medical scenarios, such as a patient on-site medical treatment scenario, a doctor remote medical treatment scenario, a patient receiving a medical examination report scenario and other diagnosis and treatment activities based on a medical examination report and the like. The client device can be used to take a picture of a paper medical examination report to obtain a medical examination report form picture, or the client device can obtain a medical examination report form picture based on a screenshot of an electronic device such as a user's mobile phone. The user can use the electronic device to transmit the medical examination report form picture to the client or the server, and the transmission manner is not limited in the embodiments of the present application. In some embodiments, each step of the medical examination report form recognition can be implemented by different implementation subjects, and the obtained results are transmitted to the next processing device. The specific form of the device is not limited in the embodiments of the present application.

[0033] In combination with Figure 2 The embodiments of the present application provide a medical examination report form recognition method, which comprises the following steps:

[0034] S1, obtaining a to-be-recognized medical examination report form picture;

[0035] S2, performing picture text comparison on the to-be-recognized medical examination report form picture to determine a large-angle tilt judgment result;

[0036] S3, rotating the to-be-recognized medical examination report form picture by a specific angle according to the large-angle tilt judgment result, so that the tilt angle of the to-be-recognized medical examination report form picture is less than 90 degrees, to obtain a first corrected picture, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees;

[0037] S4, performing tilt angle judgment on the first corrected picture to determine a tilt angle judgment result;

[0038] S5, if the tilt angle judgment result is greater than a preset angle, rotating and adjusting the first corrected picture to determine a second corrected picture; if the tilt angle judgment result is less than or equal to the preset angle, taking the first corrected picture as the second corrected picture.

[0039] S6, if the second corrected picture is a system screenshot type picture, performing table recognition on the second corrected picture through a deep learning model to determine table recognition data;

[0040] S7, if the second corrected picture is a camera shooting type picture, performing the table recognition on the second corrected picture through a text block space restoration strategy to determine the table recognition data.

[0041] The medical report table recognition method provided by the embodiment of the present application comprises the following steps: obtaining a to-be-recognized medical report table picture; performing picture text comparison on the to-be-recognized medical report table picture to determine a large-angle tilt judgment result; rotating the to-be-recognized medical report table picture by a specific angle according to the large-angle tilt judgment result, so that the tilt angle of the to-be-recognized medical report table picture is less than 90 degrees, and a first corrected picture is obtained, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees. By determining the large-angle tilt judgment result, the medical report table picture can be preliminarily adjusted in angle to prevent the situation that the medical report picture text is completely reversed or other angle offset is too large, so that the tilt angle of the medical report table picture is less than 90 degrees. The tilt angle of the first corrected picture is judged to determine a tilt angle judgment result; if the tilt angle judgment result is greater than a preset angle, the first corrected picture is rotated and adjusted to determine a second corrected picture; if the tilt angle judgment result is less than or equal to the preset angle, the first corrected picture is taken as the second corrected picture. By adjusting the first corrected picture in a small angle, the tilt angle can be less than the preset angle, so as to ensure the accuracy of subsequent table image recognition. If the second corrected picture is a system screenshot type picture, a deep learning model is used to perform table recognition on the second corrected picture to determine table recognition data; if the second corrected picture is a camera shooting type picture, a text block space restoration strategy is used to perform the table recognition on the second corrected picture to determine the table recognition data. Since the system screenshot type picture will not have paper creases, bending and other situations, the deep learning model is used to perform table recognition on the second corrected picture to improve the recognition efficiency while ensuring the recognition effect. The photographed picture is prone to recognition errors due to light, angle, paper creases and other problems, so the text block space restoration strategy is used to improve the accuracy of the medical report table recognition in the case of possible paper creases, angle distortion or light shadow. Based on this, the medical report table recognition method provided by the present application adjusts the tilt angle of the medical report table picture, and uses different recognition strategies according to the differences between the system screenshot type picture and the camera shooting type picture, so as to ensure the recognition efficiency and improve the recognition accuracy.

[0042] For step S1, an image to be recognized of a physical examination report form is acquired, where the image to be recognized of a physical examination report form refers to an image containing a physical examination report form, which can be formed by shooting or obtained by screen capture, and is not limited in the embodiment.

[0043] For step S2, image text comparison is performed on the image to be recognized of a physical examination report form to determine a large-angle inclination determination result. The image text comparison refers to text comparison on an image containing a physical examination report form. Since the text direction in a physical examination report form image should be generally consistent, and the text image is composed of fixed strokes, the database comparison can roughly determine whether the image to be recognized of a physical examination report form is obviously inclined by more than 90 degrees. For example, if the text is basically inverted, it can be preliminarily determined that the image to be recognized of a physical examination report form is inclined by about 180 degrees.

[0044] For step S3, the image to be recognized of a physical examination report form is rotated by a specific angle according to the large-angle inclination determination result, so that the inclination angle of the image to be recognized of a physical examination report form is less than 90 degrees, and a first corrected image is obtained, where the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees.

[0045] The large-angle inclination determination result can include four cases: first, the inclination angle is about 0 degrees; second, the inclination angle is about 90 degrees; third, the inclination angle is about 180 degrees; and fourth, the inclination angle is about 270 degrees. The first corrected image refers to an image obtained by rotating the image to be recognized of a physical examination report form by 0 degrees, or 90 degrees, or 180 degrees, or 270 degrees according to the large-angle inclination determination result. For example, when the inclination angle is about 0 degrees, the image to be recognized of a physical examination report form is rotated by 0 degrees. When the inclination angle is about 90 degrees, the image to be recognized of a physical examination report form is rotated by 90 degrees. When the inclination angle is about 180 degrees, the image to be recognized of a physical examination report form is rotated by 180 degrees. When the inclination angle is about 270 degrees, the image to be recognized of a physical examination report form is rotated by 270 degrees.

[0046] For step S4, an inclination angle determination is performed on the first corrected image to determine an inclination angle determination result. The inclination angle determination result refers to a refined inclination angle determination of 0-90 degrees. Unlike the above large-angle inclination determination result, the above large-angle inclination determination is a rough large-angle inclination determination result.

[0047] For step S5, if the tilt angle judgment result is greater than a preset angle, the first correction picture is adjusted by rotation to determine a second correction picture; if the tilt angle judgment result is less than or equal to the preset angle, the first correction picture is taken as the second correction picture. The preset angle is an angle considered to be set according to actual needs, indicating an acceptable tilt angle of the first correction picture. It can be 10 degrees, 20 degrees, etc., and is set according to actual needs, which is not limited in this embodiment. The tilt angle judgment result less than or equal to the preset angle indicates that the tilt angle of the first correction picture is within an acceptable range, and the rotation adjustment is not performed. The tilt angle greater than the preset angle indicates that the tilt degree of the first correction picture exceeds the acceptable range specified by human beings, and therefore the rotation adjustment is needed. The second correction picture refers to a picture meeting the preset angle specified by human beings.

[0048] In another optional embodiment of the present application, the picture text comparison on the to-be-recognized medical examination report form picture to determine a large-angle tilt judgment result comprises the steps of:

[0049] S21, obtaining standard text data;

[0050] S22, performing text comparison on the standard text data and the text in the medical examination report to determine a tilt degree between the standard text data and the text in the medical examination report; judging that the tilt degree is within a specific range to obtain the large-angle tilt judgment result, wherein the specific range is any one of 0-90 degrees, 90-180 degrees, 180-270 degrees, and 270-360 degrees;

[0051] The rotation of the to-be-recognized medical examination report form picture by 0 degrees, or 90 degrees, or 180 degrees, or 270 degrees according to the large-angle tilt judgment result to make the tilt angle of the to-be-recognized medical examination report form picture less than 90 degrees comprises the steps of:

[0052] S31, if the large-angle tilt judgment result is within 0-90 degrees, the to-be-recognized medical examination report form picture is rotated by 0 degrees;

[0053] S32, if the large-angle tilt judgment result is within 90-180 degrees, the to-be-recognized medical examination report form picture is rotated by 90 degrees;

[0054] S33, if the large-angle tilt judgment result is within 180-270 degrees, the to-be-recognized medical examination report form picture is rotated by 180 degrees;

[0055] S34, if the large-angle tilt judgment result is within 270-360 degrees, the to-be-recognized medical examination report form picture is rotated by 270 degrees.

[0056] For step S21, standard character data is obtained, which refers to a character database composed of various fonts, for comparison with the characters in the physical examination report form picture.

[0057] For step S22, the standard character data is compared with the characters in the physical examination report, and the degree of inclination between the standard character data and the characters in the physical examination report is determined; it is judged whether the degree of inclination is within a certain range to obtain the large-angle inclination judgment result, wherein the certain range is any one of 0-90 degrees, 90-180 degrees, 180-270 degrees and 270-360 degrees. The degree of inclination is a rough judgment, which can determine the inclination degree of the whole picture according to the stroke inclination degrees of multiple characters. The inclination degree judgment result is also a rough judgment, including 0-90 degrees, or 90-180 degrees, or 180-270 degrees, or 270-360 degrees.

[0058] For steps S31-S34, after judging the rough degree of inclination, the to-be-recognized physical examination report form picture is rotated by 0 degrees, 90 degrees, 180 degrees or 270 degrees to make the inclination degree of the physical examination report form picture less than 90 degrees.

[0059] Thus, the inclination degree of the physical examination report form picture is roughly adjusted to obtain the first corrected picture, which prepares for subsequent fine adjustment of the picture angle.

[0060] In another optional embodiment of the present application, the inclination angle judgment on the first corrected picture to determine the inclination angle judgment result comprises the steps of:

[0061] S41, obtaining the pixel points of the first corrected picture;

[0062] S42, selecting the pixel points with pixel values within a preset range as effective pixel points;

[0063] S43, taking one of the effective pixel points as a first effective pixel point, and determining the effective pixel point closest to the first effective pixel point as a second effective pixel point;

[0064] S44, determining an auxiliary straight line through the first effective pixel point and the second effective pixel point;

[0065] S45, if the number of effective pixel points falling on the auxiliary straight line is greater than a preset number, and the farthest distance between the effective pixel points on the auxiliary straight line is greater than a preset distance, then taking the line segment between the two effective pixel points farthest away on the auxiliary straight line as an auxiliary line segment;

[0066] S46, jump to the step of taking one of the effective pixel points as a first effective pixel point, determining the effective pixel point closest to the first effective pixel point as a second effective pixel point, and re-executing until all the effective pixel points are taken as the first effective pixel point once, to obtain a plurality of auxiliary line segments;

[0067] S47, determining the inclination angles of the plurality of auxiliary line segments;

[0068] S48, determining the inclination angle judgment result according to the distribution of the inclination angles.

[0069] For step S41, the pixel points of the first corrected picture are obtained. The pixel points refer to each pixel point constituting the first corrected picture.

[0070] For step S42, the pixel points with pixel values in a preset range are selected as effective pixel points. The preset range is a pixel value range set artificially according to actual needs, and the value of the preset range is between 0 and 255. The effective pixel points are a collection of pixel points, which refer to pixel points with pixel values in the preset range.

[0071] For step S43, one of the effective pixel points is taken as a first effective pixel point, and the effective pixel point closest to the first effective pixel point is determined as a second effective pixel point. The first effective pixel point refers to an arbitrarily selected pixel point in the effective pixel points. The second effective pixel point refers to a second effective pixel point closest to the first effective pixel point selected from the effective pixel points. The second effective pixel point can be in any direction of the first effective pixel point.

[0072] For step S44, a auxiliary straight line is determined through the first effective pixel point and the second effective pixel point. The auxiliary straight line refers to a straight line passing through the first effective pixel point and the second effective pixel point.

[0073] For step S45, if the number of effective pixel points falling on the auxiliary straight line is greater than a preset number, and the farthest distance between the effective pixel points on the auxiliary straight line is greater than a preset distance, a line segment between the two effective pixel points farthest from each other on the auxiliary straight line is taken as an auxiliary line segment. The preset number refers to a number value set artificially according to actual needs, for example, 5 pixel points or 10 pixel points, and the specific value is not limited in this embodiment. The preset distance refers to a distance value set artificially according to actual needs. The auxiliary line segment refers to a line segment formed by taking the two effective pixel points farthest from each other on the auxiliary straight line as the end points of the line segment.

[0074] For step S46, jump to the step of taking one of the effective pixel points as a first effective pixel point, determining the effective pixel point closest to the first effective pixel point as a second effective pixel point, and re-executing until all effective pixel points are taken as the first effective pixel point once, to obtain a plurality of auxiliary line segments. In this step, all effective pixel points are taken as the first pixel point, and steps S41-S45 are repeated to determine all auxiliary line segments, forming an auxiliary line segment group. The auxiliary line segment group refers to a set formed by all auxiliary line segments generated from an image.

[0075] For step S47, the inclination angle of the plurality of auxiliary line segments is determined. The inclination angle refers to the angle of inclination of the auxiliary line segment relative to the horizontal reference line in the image.

[0076] For step S48, the inclination angle determination result is determined according to the distribution of the inclination angle. The distribution of the inclination angle refers to the distribution law of the inclination angle between 0-90 degrees. According to the distribution of the inclination angle, the inclination angle determination result can be the inclination angle with the highest frequency. For example, there are 100 auxiliary line segments in total, among which 20 have an inclination angle of 10 degrees, 50 have an inclination angle of 15 degrees, 30 have an inclination angle of 20 degrees, and 20 have an inclination angle of 25 degrees. Then 15 degrees can be taken as the inclination angle determination result.

[0077] In another optional embodiment of the present application, the step of determining the inclination angle determination result according to the distribution of the inclination angle comprises the steps of:

[0078] S481, dividing 0-90 degrees into a plurality of angle intervals, and determining the number of auxiliary line segments falling within each angle interval;

[0079] S482, taking the angle interval with the most auxiliary line segments as the target angle interval;

[0080] S483, taking the average of the inclination angles of the auxiliary line segments in the target angle interval to determine the inclination angle determination result.

[0081] For step S481, 0-90 degrees is divided into a plurality of angle intervals, and the number of auxiliary line segments falling within each angle interval is determined. The angle interval refers to an interval artificially divided between 0-90 degrees. For example, 0-10 degrees is the first interval; 10-20 degrees is the second interval; 20-30 degrees is the third interval; and so on. The division can be made according to actual needs, and the uniformity of the division interval and the number of intervals are not limited in this embodiment.

[0082] For step S482, the angle interval with the most number of auxiliary line segments is taken as the target angle interval. In this step, the inclination angles of the auxiliary line segments fall within the angle interval, for example, three auxiliary line segments with inclination angles of 12 degrees, 13 degrees and 15 degrees all fall within the angle interval of 10-20 degrees. The angle interval with the most number of auxiliary line segments refers to the number of inclination angles of the auxiliary line segments that fall within the corresponding angle interval. For example, there are 3 auxiliary line segments in the angle interval of 10-20 degrees and 10 auxiliary line segments in the angle interval of 20-30 degrees, so the number of auxiliary line segments in the angle interval of 20-30 degrees is greater than that in the angle interval of 10-20 degrees. The target angle interval refers to the angle interval that contains the angle interval with the most number of auxiliary line segments.

[0083] For step S483, the inclination angles of the auxiliary line segments in the target angle interval are averaged to determine the inclination angle determination result. For example, three auxiliary line segments with inclination angles of 12 degrees, 13 degrees and 15 degrees all fall within the angle interval of 10-20 degrees, and the angle interval of 10-20 degrees has the most number of auxiliary line segments, so the inclination angle determination result is the average of 12 degrees, 14 degrees and 16 degrees, i.e., 14 degrees.

[0084] Thus, by setting auxiliary line segments and angle intervals, the inclination angle of the first corrected picture is finely adjusted to meet the recognition condition.

[0085] In another optional embodiment of the present application, if the second corrected picture is a system screenshot type picture, table recognition is performed on the second corrected picture by a deep learning model to determine table recognition data, including the steps of:

[0086] S61, pre-processing the second corrected picture to obtain a pre-processed image;

[0087] S62, extracting features of the pre-processed image using a convolutional neural network;

[0088] S63, identifying table structures in the features by a recurrent neural network model to obtain table structure data;

[0089] S64, identifying text content in the features by an optical character recognition model to obtain text content data;

[0090] S65, filling the text content data into the table structure data accordingly to obtain the table recognition data.

[0091] For step S61, the second corrected picture is preprocessed to obtain a preprocessed image. The preprocessing includes but is not limited to adjusting the size of the image, cropping, standardization, etc., to ensure the consistency and quality of the input image.

[0092] For step S62, the features of the preprocessed image are extracted using a convolutional neural network. A convolutional neural network (CNN) is used to extract the features of the image. CNN can learn the local and global features of the image, helping the model understand the structure and content of the table.

[0093] For step S63, the table structure in the features is identified by a recurrent neural network model to obtain table structure data. The recurrent neural network (RNN) is a type of recursive neural network that takes sequence data as input and performs recursion in the evolution direction of the sequence. All nodes (recurrent units) are connected in a chain. The table structure refers to the elements of the table, such as rows, columns, titles, and their positions and relationships.

[0094] For step S64, the text content in the features is identified by an optical character recognition model to obtain text content data. The optical character recognition model (OCR) refers to analyzing and processing image files to obtain text and layout information. The text content data refers to the recognized text content.

[0095] For step S65, the text content data is filled into the table structure data accordingly to obtain the table recognition data. This step specifically refers to filling the correct text content into the corresponding table cells according to the structure of the table and the text recognition results.

[0096] Thus, the table recognition data is determined by the deep learning model for table recognition, accurately and efficiently recognizing the physical examination report table information in the picture.

[0097] In another optional embodiment of the present application, the table recognition of the second corrected picture by the text block space restoration strategy to determine the table recognition data includes the following steps:

[0098] S71, performing optical character recognition on the second corrected picture to obtain text data;

[0099] S72, determining the connected phrases in the text data to obtain text blocks, the text blocks including text block content and text block coordinates, wherein the text block coordinates are any vertex coordinates of the rectangular space where the text blocks are located;

[0100] S73, input the text block content into a text attribute judgment model to determine a table header information text block, a first column content text block, and a table content information text block;

[0101] S74, fill the table header information text block into a table header row of a table, fill the first column content text block into a first column of the table, and fill the text content into the table according to the text block coordinates of the table content information text block to obtain table recognition data.

[0102] For step S71, optical character recognition is performed on the second corrected picture to obtain text data. The optical character recognition refers to recognizing characters in the picture to obtain text data information.

[0103] For step S72, connected phrases in the text data are determined to obtain text blocks, and the text blocks include text block content and text block coordinates. The text block coordinates are any vertex coordinates of a rectangular space where the text block is located. The coordinates are coordinate data set according to a pixel point matrix of the picture. The rectangular space is a rectangle labeled according to the text block. The any point coordinates can be any one of four vertexes of the rectangular space, but all the text blocks must be consistent. For example, all the text blocks select a vertex at the upper left corner of the rectangular space, or all the text blocks select a vertex at the lower right corner of the rectangular space.

[0104] For step S73, the text content is input into a text attribute judgment model to determine a table header information text block, a first column content text block, and a table content information text block. The text attribute judgment model classifies the text content based on the text content. First, table header information is distinguished from table content information. For example, texts such as “physical examination item” and “examination result” are the table header information text block, texts such as “red blood cell count” and “white blood cell count” are the first column content text block, and texts such as “0.34” and “um / L” are the table content information text block. In the specific implementation process, the recognized text content is subjected to phrase similarity calculation with the existing content in a word table library. The calculation can include various similarity calculation algorithms, such as Hamming distance and edit distance, which are not specifically limited in the embodiment. Through the similarity calculation, phrases with a similarity satisfying a preset condition are obtained. According to a preset phrase category, it is determined that the text content belongs to the table header information text block, the first column content text block, or the table content information text block.

[0105] For step S74, fill the table header information text block into the table header row, fill the first column content text block into the first column of the table, fill the text content into the table according to the text block coordinates of the table content information text block, and obtain the table recognition data. According to the text block coordinates of the table header row and the first column and the text block coordinates of the table content information text block, the position of the table content information text block relative to the table header row and the first column can be determined, and the text content is filled into the table to obtain the table recognition data.

[0106] In another optional embodiment of the present application, the step of filling the text content into the table according to the text block coordinates of the table content information text block to obtain the table recognition data comprises the steps of:

[0107] S741, comparing the horizontal coordinate of any one of the table content information text blocks with the horizontal coordinates of all the first column content text blocks to determine the row where the table content information text block is located;

[0108] S742, comparing the vertical coordinate of any one of the table content information text blocks with the vertical coordinates of all the table header information text blocks to determine the column where the table content information text block is located;

[0109] S743, filling the text block content in the table content information text block into the table according to the row and the column where the table content information text block is located;

[0110] S744, traversing all the table content information text blocks until the table is filled completely to form the table recognition data.

[0111] For step S741, compare the horizontal coordinate of any one of the table content information text blocks with the horizontal coordinates of all the first column content text blocks to determine the row where the table content information text block is located. Compare the horizontal coordinate of any one of the table content information text blocks with the horizontal coordinates of all the first column content text blocks, select the row where the first column content text block with the horizontal coordinate closest to the horizontal coordinate of the table content information text block is located, and the row is the row where the table content information text block is located. The comparison method can be difference or ratio, which is not limited in the embodiment.

[0112] For step S742, the ordinate of any one of the table content text blocks is compared with the ordinates of all the table header text blocks to determine the column containing the table content text block. The column containing the table header text block whose ordinate is closest to the ordinate of the table content text block is selected as the column containing the table content text block. The comparison method can be subtraction, ratio, etc., and is not specifically limited in this embodiment.

[0113] For step S743, based on the row and column of the table content information text block, the text block content is filled into the table; S744, all the table content information text blocks are traversed until the table is completely filled, forming the table recognition data. That is, based on the row and column, the text block content is filled into the corresponding row and column positions of the table until the table is completely filled, forming the table recognition data.

[0114] Therefore, by comparing the horizontal and vertical coordinates between text blocks, the table is filled, avoiding the inaccurate recognition problems caused by curled or wrinkled paper in the image. This improves the accuracy of recognition.

[0115] Combination Figure 3 As shown, in one embodiment, the present invention provides a medical examination report form recognition device, which corresponds one-to-one with the medical examination report form recognition method in the above embodiments. The medical examination report form recognition device includes an acquisition module 101, a correction module 102, and a processing module 103. Detailed descriptions of each functional module are as follows:

[0116] Module 101 is used to acquire the image of the physical examination report form to be identified;

[0117] Correction module 102 is used to perform image-text comparison on the image of the physical examination report form to be identified, and determine the large-angle tilt judgment result;

[0118] The correction module 102 is further configured to rotate the physical examination report form image to be identified by a specific angle according to the large angle tilt judgment result, so that the tilt angle of the physical examination report form image to be identified is less than 90 degrees, and obtain a first corrected image, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees;

[0119] The correction module 102 is also used to determine the tilt angle of the first corrected image and determine the tilt angle determination result.

[0120] The correction module 102 is further configured to, if the tilt angle determination result is greater than a preset angle, rotate and adjust the first correction image to determine the second correction image; if the tilt angle determination result is less than or equal to the preset angle, use the first correction image as the second correction image.

[0121] Processing module 103 is used to perform table recognition on the second corrected image using a deep learning model if the second corrected image is a system screenshot type image, and determine the table recognition data.

[0122] The processing module 103 is further configured to, if the second corrected image is a camera-captured image, perform the table recognition on the second corrected image using a text block space restoration strategy to determine the table recognition data.

[0123] In one embodiment, the correction module 102 is specifically used for:

[0124] Obtain standard text data;

[0125] The standard text data is compared with the text in the medical examination report to determine the degree of tilt between the standard text data and the text in the medical examination report; the degree of tilt is determined to be within a specific range to obtain the large-angle tilt judgment result, wherein the specific range is any one of 0-90 degrees, 90-180 degrees, 180-270 degrees and 270-360 degrees;

[0126] If the large-angle tilt judgment result is between 0 and 90 degrees, then rotate the physical examination report form image to be identified by 0 degrees;

[0127] If the large-angle tilt judgment result is between 90 and 180 degrees, then rotate the physical examination report form image to be identified by 90 degrees;

[0128] If the large-angle tilt judgment result is between 180 and 270 degrees, then rotate the physical examination report form image to be identified by 180 degrees;

[0129] If the large-angle tilt judgment result is between 270 and 360 degrees, then the physical examination report form image to be identified is rotated by 270 degrees.

[0130] In one embodiment, the correction module 102 is specifically used for:

[0131] Obtain the pixels of the first corrected image;

[0132] Pixels whose pixel values ​​are within a preset range are selected as valid pixels;

[0133] One of the valid pixels is designated as the first valid pixel, and the valid pixel closest to the first valid pixel is designated as the second valid pixel.

[0134] An auxiliary straight line is determined using the first valid pixel and the second valid pixel;

[0135] If the number of effective pixels falling on the auxiliary line is greater than a preset number, and the farthest distance between the effective pixels on the auxiliary line is greater than a preset distance, then the line segment between the two farthest effective pixels on the auxiliary line is taken as an auxiliary line segment.

[0136] Jump to the step of taking one of the valid pixels as the first valid pixel, determining the valid pixel closest to the first valid pixel as the second valid pixel, and repeating the process until all valid pixels have been taken as the first valid pixel once, thus obtaining multiple auxiliary line segments;

[0137] Determine the tilt angle of the multiple auxiliary line segments;

[0138] The tilt angle judgment result is determined based on the distribution of the tilt angles.

[0139] In one embodiment, the correction module 102 is specifically used for:

[0140] Divide the 0-90 degree range into multiple angle intervals and determine the number of auxiliary line segments that the tilt angle falls within each angle interval.

[0141] The angle interval with the largest number of auxiliary line segments is taken as the target angle interval;

[0142] The tilt angle judgment result is determined by averaging the tilt angles of the auxiliary line segments within the target angle range.

[0143] In one embodiment, the processing module 103 is specifically used for:

[0144] The second corrected image is preprocessed to obtain the preprocessed image;

[0145] The features of the preprocessed image are extracted using a convolutional neural network;

[0146] The table structure data is obtained by identifying the table structure in the features using a recurrent neural network model.

[0147] The text content data is obtained by recognizing the features using an optical character recognition model.

[0148] The text content data is filled into the table structure data accordingly to obtain the table recognition data.

[0149] In one embodiment, the processing module 103 is specifically used for:

[0150] Optical character recognition is performed on the second corrected image to obtain text data;

[0151] Identify the connected phrases in the text data to obtain a text block. The text block includes text block content and text block coordinates, wherein the text block coordinates are the coordinates of any vertex in the rectangular space where the text block is located.

[0152] The text block content is input into the text attribute judgment model to determine the header information text block, the first column content text block, and the table content information text block.

[0153] The header information text block is filled into the header row of the table, and the first column content text block is filled into the first column of the table. Based on the text block coordinates of the table content information text block, the text content is filled into the table to obtain the table recognition data.

[0154] Based on the coordinates of the text block in the table content information text block, the text content is filled into the table to obtain the table recognition data.

[0155] In one embodiment, the processing module 103 is specifically used for:

[0156] The horizontal coordinate of any one of the table content information text blocks is compared with the horizontal coordinates of all the first column content text blocks to determine the row where the table content information text block is located.

[0157] The column of the table content information text block is determined by comparing the vertical coordinate of any one of the table content information text blocks with the vertical coordinates of all the table header information text blocks.

[0158] Traverse all the text blocks containing the table content information until the table is completely filled, thus forming the table recognition data.

[0159] This invention provides a medical examination report form recognition device. It acquires an image of a medical examination report form to be recognized; performs image-text comparison on the image to determine a large-angle tilt judgment result; and rotates the image to be recognized by a specific angle based on the large-angle tilt judgment result, so that the tilt angle of the image is less than 90 degrees, obtaining a first corrected image. The specific angle can be any one of 0 degrees, 90 degrees, 180 degrees, and 270 degrees. By determining the large-angle tilt judgment result, a preliminary angle adjustment can be made to the medical examination report form image to prevent the text from being completely inverted or other angles from being excessively offset, ensuring that the tilt angle of the image is less than 90 degrees. The first corrected image is then subjected to a tilt angle judgment to determine a tilt angle judgment result. If the tilt angle judgment result is greater than a preset angle, the first corrected image is rotated to determine a second corrected image; if the tilt angle judgment result is less than or equal to the preset angle, the first corrected image is used as the second corrected image. By adjusting the first corrected image at a small angle, the tilt angle can be made smaller than a preset angle, ensuring the accuracy of subsequent table image recognition. If the second corrected image is a system screenshot, a deep learning model is used to perform table recognition on the second corrected image to determine the table recognition data; if the second corrected image is a camera-captured image, a text block space restoration strategy is used to perform table recognition on the second corrected image to determine the table recognition data. Since system screenshots do not have paper wrinkles or bends, using a deep learning model to perform table recognition on the second corrected image can improve recognition efficiency while ensuring recognition effect. However, captured images are prone to recognition errors due to lighting, angle, paper wrinkles, etc. Therefore, the text block space restoration strategy improves the accuracy of the medical examination report table recognition even when paper wrinkles, angular distortion, or shadows are present. Based on this, the medical examination report table recognition device provided by this invention adjusts the tilt angle of the medical examination report table image and adopts different recognition strategies according to system screenshots and camera-captured images, ensuring both recognition efficiency and improving recognition accuracy.

[0160] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a physical examination report form recognition method on the server side.

[0161] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements the client-side functions or steps of a medical examination report form recognition method.

[0162] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0163] S1, Obtain the image of the medical examination report form to be identified;

[0164] S2, perform image-text comparison on the image of the physical examination report form to be identified, and determine the large-angle tilt judgment result;

[0165] S3, based on the large-angle tilt judgment result, rotate the physical examination report form image to be identified by a specific angle so that the tilt angle of the physical examination report form image to be identified is less than 90 degrees, to obtain the first corrected image, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees;

[0166] S4, determine the tilt angle of the first corrected image and determine the tilt angle judgment result;

[0167] S5, if the tilt angle determination result is greater than the preset angle, then the first corrected image is rotated and adjusted to determine the second corrected image; if the tilt angle determination result is less than or equal to the preset angle, then the first corrected image is used as the second corrected image.

[0168] S6, If the second corrected image is a system screenshot, perform table recognition on the second corrected image using a deep learning model to determine the table recognition data;

[0169] S7. If the second corrected image is a camera-captured image, the table recognition is performed on the second corrected image using a text block space restoration strategy to determine the table recognition data.

[0170] The computer device provided in this embodiment of the invention has the same advantages over the prior art as the above-mentioned physical examination report form recognition method and physical examination report form recognition device, and will not be repeated here.

[0171] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0172] S1, Obtain the image of the medical examination report form to be identified;

[0173] S2, perform image-text comparison on the image of the physical examination report form to be identified, and determine the large-angle tilt judgment result;

[0174] S3, based on the large-angle tilt judgment result, rotate the physical examination report form image to be identified by a specific angle so that the tilt angle of the physical examination report form image to be identified is less than 90 degrees, to obtain the first corrected image, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees;

[0175] S4, determine the tilt angle of the first corrected image and determine the tilt angle judgment result;

[0176] S5, if the tilt angle determination result is greater than the preset angle, then the first corrected image is rotated and adjusted to determine the second corrected image; if the tilt angle determination result is less than or equal to the preset angle, then the first corrected image is used as the second corrected image.

[0177] S6, If the second corrected image is a system screenshot, perform table recognition on the second corrected image using a deep learning model to determine the table recognition data;

[0178] S7. If the second corrected image is a camera-captured image, the table recognition is performed on the second corrected image using a text block space restoration strategy to determine the table recognition data.

[0179] The computer-readable storage medium provided in this embodiment of the invention has the same advantages over the prior art as the above-mentioned physical examination report form recognition method and physical examination report form recognition device, and will not be repeated here.

[0180] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A medical examination report form identification method, characterized by, The method comprises the following steps: acquiring a picture of a physical examination report form to be recognized; performing picture-text comparison on the picture of the physical examination report form to be recognized to determine a large-angle inclination determination result; rotating the picture of the physical examination report form to be recognized by a specific angle according to the large-angle inclination determination result, so that the inclination angle of the picture of the physical examination report form to be recognized is less than 90 degrees, to obtain a first corrected picture, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees; performing inclination angle determination on the first corrected picture to determine an inclination angle determination result; if the inclination angle determination result is greater than a preset angle, performing rotation adjustment on the first corrected picture to determine a second corrected picture; if the inclination angle determination result is less than or equal to the preset angle, taking the first corrected picture as the second corrected picture; if the second corrected picture is a picture of a system screenshot type, performing table recognition on the second corrected picture by a deep learning model to determine table recognition data; if the second corrected picture is a picture of a camera shooting type, performing the table recognition on the second corrected picture by a text block space restoration strategy to determine the table recognition data; performing optical character recognition on the second corrected picture to obtain text data; determining a connected phrase in the text data to obtain a text block, wherein the text block comprises text block content and text block coordinates, and the text block coordinates are any vertex coordinates of a rectangular space where the text block is located; inputting the text block content into a text attribute determination model to determine a table header information text block, a first column content text block and table content information text blocks; filling the table header information text block into a table header row of a table, filling the first column content text block into a first column of the table, and filling the text content of the table content information text blocks into the table according to the text block coordinates of the table content information text blocks to obtain the table recognition data; comparing the horizontal coordinates of any one of the table content information text blocks with the horizontal coordinates of all the first column content text blocks to determine the row where the table content information text block is located; comparing the vertical coordinates of any one of the table content information text blocks with the vertical coordinates of all the table header information text blocks to determine the column where the table content information text block is located; filling the text block content in the table content information text block into the table according to the row and the column where the table content information text block is located; iterating through all the table content information text blocks until the table is filled completely to form the table recognition data.

2. The medical report form recognition method according to claim 1, wherein The method of performing picture-text comparison on the picture of the physical examination report form to be recognized to determine a large-angle inclination determination result comprises the following steps: acquiring standard text data; performing text comparison between the standard text data and the text in the physical examination report to determine the inclination degree between the standard text data and the text in the physical examination report; and determining the large-angle inclination determination result if the inclination degree is within a specific range, wherein the specific range is any one of the ranges of 0-90 degrees, 90-180 degrees, 180-270 degrees and 270-360 degrees. The first correction picture is rotated 0 degrees, or 90 degrees, or 180 degrees, or 270 degrees according to the large-angle inclination judgment result, so that the inclination angle of the to-be-recognized medical report form picture is less than 90 degrees, and the inclination angle judgment result is determined. If the large-angle inclination judgment result is between 0-90 degrees, the to-be-recognized medical report form picture is rotated 0 degrees. If the large-angle inclination judgment result is between 90-180 degrees, the to-be-recognized medical report form picture is rotated 90 degrees. If the large-angle inclination judgment result is between 180-270 degrees, the to-be-recognized medical report form picture is rotated 180 degrees. If the large-angle inclination judgment result is between 270-360 degrees, the to-be-recognized medical report form picture is rotated 270 degrees.

3. The medical report form recognition method of claim 1, wherein, The inclination angle of the first correction picture is judged, and an inclination angle judgment result is determined, including: Obtaining the pixel points of the first correction picture; Selecting the pixel points with pixel values in a preset range as effective pixel points; Taking one of the effective pixel points as a first effective pixel point, determining the effective pixel point closest to the first effective pixel point as a second effective pixel point; Determining an auxiliary straight line through the first effective pixel point and the second effective pixel point; If the number of effective pixel points falling on the auxiliary straight line is greater than a preset number, and the maximum distance between the effective pixel points on the auxiliary straight line is greater than a preset distance, then the line segment between the two effective pixel points farthest away on the auxiliary straight line is taken as an auxiliary line segment; Jumping to the step of taking one of the effective pixel points as a first effective pixel point, and determining the effective pixel point closest to the first effective pixel point as a second effective pixel point, and re-executing until all effective pixel points have been taken as the first effective pixel point once, to obtain a plurality of auxiliary line segments; Determining the inclination angles of the plurality of auxiliary line segments; According to the distribution of the inclination angles, the inclination angle judgment result is determined.

4. The medical report form recognition method according to claim 3, wherein According to the distribution of the inclination angles, the inclination angle judgment result is determined, including: Dividing 0-90 degrees into a plurality of angle intervals, and determining the number of auxiliary line segments falling in each angle interval; Taking the angle interval with the largest number of auxiliary line segments as a target angle interval; Taking the average of the inclination angles of the auxiliary line segments in the target angle interval to determine the inclination angle judgment result.

5. The method of claim 1, wherein, If the second correction picture is a system screenshot type picture, a deep learning model is used to perform table recognition on the second correction picture to determine table recognition data, including: Preprocessing the second correction picture to obtain a preprocessed image; Using a convolutional neural network to extract features of the preprocessed image; Identifying the table structure in the features through a recurrent neural network model to obtain table structure data; Identifying the text content in the features through an optical character recognition model to obtain text content data; The text content data is filled into the table structure data correspondingly to obtain the table recognition data.

6. A physical examination report form recognition apparatus characterized by comprising: The method comprises the following steps: The acquisition module is configured to acquire a picture of a physical examination report table to be recognized. The correction module is configured to perform picture-text comparison on the picture of the physical examination report table to be recognized to determine a large-angle tilt determination result. The correction module is further configured to rotate the picture of the physical examination report table to be recognized by a specific angle according to the large-angle tilt determination result, so that the tilt angle of the picture of the physical examination report table to be recognized is less than 90 degrees, to obtain a first corrected picture, wherein the specific angle is any one of 0 degrees, 90 degrees, 180 degrees and 270 degrees. The correction module is further configured to perform tilt angle determination on the first corrected picture to determine a tilt angle determination result. The correction module is further configured to perform rotation adjustment on the first corrected picture to determine a second corrected picture if the tilt angle determination result is greater than a preset angle, or to take the first corrected picture as the second corrected picture if the tilt angle determination result is less than or equal to the preset angle. The processing module is configured to perform table recognition on the second corrected picture by a deep learning model if the second corrected picture is a system screenshot type picture, to determine table recognition data. The processing module is further configured to perform the table recognition on the second corrected picture by a text block space restoration strategy if the second corrected picture is a camera shooting type picture, to determine the table recognition data. The second corrected picture is subjected to optical character recognition to obtain text data. A connected phrase in the text data is determined to obtain a text block, wherein the text block comprises text block content and text block coordinates, and the text block coordinates are any vertex coordinates of a rectangular space where the text block is located. The text block content is input into a text attribute determination model to determine a table header information text block, a first column content text block and table content information text blocks. The table header information text block is filled into a table header row of a table, the first column content text block is filled into a first column of the table, and the text content of the table content information text blocks is filled into the table according to the text block coordinates of the table content information text blocks, to obtain the table recognition data. The horizontal coordinates of any one of the table content information text blocks are compared with the horizontal coordinates of all the first column content text blocks to determine a row where the table content information text blocks are located. The vertical coordinates of any one of the table content information text blocks are compared with the vertical coordinates of all the table header information text blocks to determine a column where the table content information text blocks are located. The text block content in the table content information text blocks is filled into a table according to the row and the column where the table content information text blocks are located. All the table content information text blocks are traversed until the table is completely filled to form the table recognition data.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the physical examination report table recognition method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, wherein the computer program comprises the following steps of: The computer program, when executed by a processor, implements the steps of the physical examination report form recognition method according to any one of claims 1 to 5.

Images