Clearing receipt data identification system and method thereof

Abstract

The invention relates to the technical field of receipt data identification, and discloses a liquidation receipt data identification system and method, and the system comprises an image obtaining and preprocessing module which is used for shooting five images of a target receipt from multiple angles, carrying out the alignment, carrying out the fusion of the aligned images, carrying out the noise reduction and contrast enhancement processing, and carrying out the recognition of the five images; obtaining a first feature image; the image recognition module is used for extracting receipt data from the first feature image; the field identification and classification module is used for identifying contents and meanings of fields in the receipt data and classifying the fields; the field association analysis module is used for identifying a logical relationship between fields in the receipt data and matching the fields in the receipt data; the data cleaning and verifying module is used for cleaning the matched receipt data and verifying the content of the matched receipt data; and the data output module is used for carrying out structured output on the cleaned and verified receipt data. According to the invention, the accuracy of receipt data identification is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of receipt data recognition, and in particular relates to a clearing receipt data recognition system and method. Background Art

[0002] With the rapid development of intelligent data processing technology, automatic receipt recognition has gradually become an important tool for enterprises to efficiently process financial and reimbursement data. At present, receipt recognition systems based on optical character recognition (OCR) have been widely used in receipt data extraction. However, existing OCR technology still faces many limitations in practical applications, especially when processing complex receipt images. First, image quality significantly affects the accuracy of OCR recognition. Since receipts usually have problems such as blurry shooting, tilt, and uneven lighting, the OCR system is easily affected during recognition, resulting in an increase in recognition error rate. For example, blurred characters or low resolution in the image can make it difficult for OCR to accurately read the content, while reflections and shadows can also block key fields. Secondly, another key defect of existing OCR technology is the lack of contextual understanding. OCR is mainly used to recognize characters word by word, lacks semantic level analysis, and cannot understand the logical relationship between fields. For example, for important information such as date, amount, and product name in the receipt, OCR may recognize these characters, but cannot understand the meaning and relationship between these fields. This defect makes the system prone to data mismatch, especially when there are no standardized labels or the field positions are not fixed, which further affects the structured processing of the data. Summary of the invention

[0003] The present invention provides a clearing receipt data recognition system and method thereof, which solve the technical problems in the related art that the recognition accuracy is low and data mismatch and erroneous recognition are prone to occur due to limitations of image quality and lack of context understanding.

[0004] The present invention provides a clearing receipt data recognition system, including an image acquisition and preprocessing module, an image recognition module, a field recognition and classification module, a field association analysis module, a data cleaning and verification module and a data output module: The image acquisition and preprocessing module is used to take five images of the target receipt from multiple angles and align them using image registration technology, specifically as follows: take images from the front, upper left, lower left, upper right and lower right angles of the target receipt, identify feature points in each image using a scale-invariant feature transformation algorithm, output the coordinates and feature descriptors of the feature points, calculate the similarity between the feature points using Euclidean distance, match the feature points, eliminate mismatches by calculating the distance ratio, and align similar areas of the five images using affine transformation; The steps of using the scale-invariant feature transformation algorithm include: Step S201, generating Gaussian difference images of different scales based on five images at different angles; Step S202, finding potential feature points from the Gaussian difference image by extreme value detection, filtering edge responses and removing low contrast of the potential feature points to obtain feature points, and extracting the coordinates of the feature points; Step S203, assigning a main direction to each feature point by calculating the gradient direction histogram of the feature point neighborhood; Step S204, concatenating the gradient direction histograms of the feature point neighborhood to obtain a feature descriptor of the feature point; The distance ratio is calculated as follows: ,in, represents the distance ratio, Indicates the distance between the current feature point descriptor and the nearest feature point. Indicates the distance between the current feature point descriptor and the next closest feature point; when the distance ratio is greater than the third threshold, the match is a false match and needs to be removed. The third threshold is a custom parameter; The step of aligning similar regions of the five images using affine transformation comprises: Step S301, the image taken at the front angle is used as the reference image, and another image is selected at the same time, K pairs of matching feature points are found in the two images, and an affine transformation matrix is ​​constructed based on the coordinate transformation formula of the affine transformation principle and the matching feature points. The matrix expression is: , where T represents the affine transformation matrix, a, b, c and d represent the first coefficient, the second coefficient, the third coefficient and the fourth coefficient respectively, which are used to control the scaling, rotation and tilt of the image, tx and ty represent the first translation parameter and the second translation parameter respectively, and the coordinate transformation formula of the affine transformation principle is: ,in and are the horizontal and vertical coordinates of the pixel coordinates after transformation, respectively. x and y are the horizontal and vertical coordinates of the pixel coordinates before transformation, respectively. a, b, c, d, tx and ty are solved by solving a linear equation system constructed by matching feature points and combining coordinate transformation formula. Step S302, using a random sampling consistency algorithm to filter out mismatched feature points; Step S303, using an affine transformation matrix, according to the coordinate transformation formula in step S301, the pixel coordinates of another image are calculated and transformed through the pixel coordinates of the reference image; Step S304, repeating steps S301 to S303, applying the transformation to all images except the reference image to align them with the reference image; Applying a fusion algorithm to fuse the aligned images to obtain a fused image, performing noise reduction and contrast enhancement processing on the fused image to obtain a first feature image; An image recognition module, configured to extract receipt data from the first feature image using image recognition technology; The field recognition and classification module is used to recognize the content and meaning of the fields in the receipt data and classify the fields, including: extracting all text blocks from the first feature image using optical character recognition technology to obtain receipt data, wherein each text block contains a group of continuous characters; A field association analysis module, used to identify the logical relationship between the fields in the receipt data using a dependency syntax analysis algorithm, and match the fields in the receipt data; Data cleaning and verification module, used to clean and verify the content of the matched receipt data; The data output module is used to output the cleaned and verified receipt data in a structured manner.

[0005] Furthermore, the aligned images are fused using a fusion algorithm, including: Step S401, defining a neighborhood window of size N×N around each pixel of each image, counting the frequency distribution of each gray level in the neighborhood window, and generating a gray level histogram of the neighborhood window, wherein N is a custom parameter and the gray level represents the brightness of the pixel in the image; Step S402, calculating the entropy value of the pixel point according to the grayscale histogram, and selecting the pixel with the highest entropy value in the same pixel position in the five images. The entropy value calculation formula is: , where H represents the entropy value of the pixel, represents the frequency of gray level i appearing in the neighborhood window, and J represents the number of gray levels; Step S403, filling the pixel with the maximum entropy value at each pixel position into the corresponding position of the final fused image to generate a complete fused image.

[0006] Furthermore, the characters in each text block are identified to obtain a text string, and the identified text string is semantically parsed using natural language processing technology to extract field content and meaning, and the identified field content is matched with the rules according to a predefined rule base to determine the category of the field, wherein the rule base contains field names and variants of field names.

[0007] Furthermore, a transfer-based dependency parsing algorithm is used to analyze the logical relationship between fields in the receipt data. The specific steps include: first, converting the field content into an input format suitable for dependency parsing, then building a dependency tree by applying the LEFT-ARC transfer operation to form a dependency graph between fields, and then identifying the logical relationship between fields. Finally, matching the field content is completed according to the dependency graph.

[0008] Furthermore, the matched receipt data is cleaned and content verified. Data cleaning includes: formatting the field content, removing spaces and redundant characters, and spelling correction; content verification includes: field format verification, logical relationship verification and consistency check.

[0009] The present invention provides a method for identifying clearing receipt data, comprising the following steps: Step S601, taking five images of the target receipt from multiple angles, aligning them using image registration technology, fusing the aligned images using a fusion algorithm to obtain a fused image, and performing noise reduction and contrast enhancement on the fused image; Step S602, extracting receipt data from the first feature image using image recognition technology; Step S603, identifying the content and meaning of the fields in the receipt data, and classifying the fields; Step S604, using a dependency parsing algorithm to identify the logical relationship between the fields in the receipt data, and matching the fields in the receipt data; Step S605, cleaning and content verification of the matched receipt data; Step S606: Output the cleaned and verified receipt data in a structured manner.

[0010] The beneficial effects of the present invention are as follows: the present invention effectively reduces recognition errors caused by problems such as uneven lighting and tilt, and improves image clarity and contrast by photographing receipts from multiple angles and performing image registration and fusion processing; The present invention uses feature point matching algorithm and affine transformation to align images, so that the key information of multi-angle images is consistent, thereby ensuring the accuracy of image fusion and data consistency; The present invention uses natural language processing technology to perform word segmentation and entity recognition on the recognized text, and combines it with a rule library to match field content, which can accurately distinguish and classify the fields in the receipt, effectively improving the accuracy of data classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a module schematic diagram of a clearing receipt data recognition system of the present invention; Figure 2is a flow chart of the scale-invariant feature transformation algorithm of the present invention; Figure 3 It is a flow chart of using the fusion algorithm of the present invention.

[0012] In the figure: image acquisition and preprocessing module 101, image recognition module 102, field recognition and classification module 103, field association analysis module 104, data cleaning and verification module 105 and data output module 106. DETAILED DESCRIPTION

[0013] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0014] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0015] like Figure 1-Figure 3 As shown, a clearing receipt data recognition system includes an image acquisition and preprocessing module, an image recognition module, a field recognition and classification module, a field association analysis module, a data cleaning and verification module and a data output module: The image acquisition and preprocessing module 101 is used to take five images of the target receipt from multiple angles and align them using image registration technology, specifically as follows: take images from the front, upper left, lower left, upper right and lower right angles of the target receipt respectively, identify feature points in each image using a scale-invariant feature transformation algorithm, output the coordinates and feature descriptors of the feature points, calculate the similarity between the feature points using Euclidean distance, match the feature points, eliminate mismatches by calculating the distance ratio, and align similar areas of the five images using affine transformation; The steps of using the scale-invariant feature transformation algorithm include: Step S201, generating Gaussian difference images of different scales based on five images at different angles; Step S202, finding potential feature points from the Gaussian difference image by extreme value detection, filtering edge responses and removing low contrast of the potential feature points to obtain feature points, and extracting the coordinates of the feature points; Step S203, assigning a main direction to each feature point by calculating the gradient direction histogram of the feature point neighborhood; Step S204, concatenating the gradient direction histograms of the feature point neighborhood to obtain a feature descriptor of the feature point; The distance ratio is calculated as follows: ,in, represents the distance ratio, Indicates the distance between the current feature point descriptor and the nearest feature point. Indicates the distance between the current feature point descriptor and the next closest feature point; when the distance ratio is greater than the third threshold, the match is a false match and needs to be removed. The third threshold is a custom parameter; The step of aligning similar regions of the five images using affine transformation comprises: Step S301, the image taken at the front angle is used as the reference image, and another image is selected at the same time, K pairs of matching feature points are found in the two images, and an affine transformation matrix is ​​constructed based on the coordinate transformation formula of the affine transformation principle and the matching feature points. The matrix expression is: , where T represents the affine transformation matrix, a, b, c and d represent the first coefficient, the second coefficient, the third coefficient and the fourth coefficient respectively, which are used to control the scaling, rotation and tilt of the image, tx and ty represent the first translation parameter and the second translation parameter respectively, and the coordinate transformation formula of the affine transformation principle is: ,in and are the horizontal and vertical coordinates of the pixel coordinates after transformation, respectively, x and y are the horizontal and vertical coordinates of the pixel coordinates before transformation, respectively, a, b, c, d, tx and ty are obtained by solving a linear equation system constructed by matching feature points and combining coordinate transformation formulas, preferably, K is set to 3; Step S302, using a random sampling consistency algorithm to filter out mismatched feature points; Step S303, using an affine transformation matrix, according to the coordinate transformation formula in step S301, the pixel coordinates of another image are calculated and transformed through the pixel coordinates of the reference image; Step S304, repeating steps S301 to S303, applying the transformation to all images except the reference image to align them with the reference image; Applying a fusion algorithm to fuse the aligned images to obtain a fused image, performing noise reduction and contrast enhancement processing on the fused image to obtain a first feature image; An image recognition module 102, configured to extract receipt data from the first feature image using image recognition technology; The field recognition and classification module 103 is used to recognize the content and meaning of the fields in the receipt data and classify the fields, including: extracting all text blocks from the first feature image using optical character recognition technology to obtain receipt data, wherein each text block contains a group of continuous characters; A field association analysis module 104, for identifying the logical relationship between the fields in the receipt data using a dependency syntax analysis algorithm, and matching the fields in the receipt data; Data cleaning and verification module 105, used to clean and verify the content of the matched receipt data; The data output module 106 is used to output the cleaned and verified receipt data in a structured manner.

[0016] In one embodiment of the present invention, the specific steps of step S201 include: Step S2011, performing Gaussian blur processing on the five images; Step S2012, gradually reducing the image after Gaussian blur processing to generate an image pyramid, that is, images of different scales; Step S2013, within each scale, the difference between images of adjacent blur levels is calculated to form a Gaussian difference image.

[0017] In one embodiment of the present invention, the specific steps of step S202 include: Step S2021, in each Gaussian difference image, the value of each pixel is compared with the values ​​of the pixels at the current scale and the adjacent scales, and when the value of the pixel is a maximum value or a minimum value, it is marked as a potential feature point; Step S2022, calculating the curvature of the potential feature point, and filtering out the potential feature point when the curvature of the potential feature point is greater than a first threshold value, wherein the curvature of the potential feature point The calculation formula is: ,in represents the secondary partial derivative of the differential Gaussian image in the x direction, represents the secondary partial derivative of the differential Gaussian image in the y direction, represents the mixed partial derivatives of the differential Gaussian image in the x and y directions, preferably, the first threshold is 10; Step S2023, calculating the contrast of the potential feature point, when the contrast of the potential feature point is less than a second threshold, removing the potential feature point, preferably, the second threshold is 0.03.

[0018] In one embodiment of the present invention, the aligned images are fused using a fusion algorithm, including: Step S401, defining a neighborhood window of size N×N around each pixel of each image, counting the frequency distribution of each gray level in the neighborhood window, and generating a gray level histogram of the neighborhood window, wherein N is a custom parameter, preferably, N is set to 5, and the gray level represents the brightness of the pixel in the image; Step S402, calculating the entropy value of the pixel point according to the grayscale histogram, and selecting the pixel with the highest entropy value in the same pixel position in the five images. The entropy value calculation formula is: , where H represents the entropy value of the pixel, represents the frequency of gray level i appearing in the neighborhood window, J represents the number of gray levels, and for 8-bit gray level, the number of gray levels is 256; Step S403, filling the pixel with the maximum entropy value at each pixel position into the corresponding position of the final fused image to generate a complete fused image.

[0019] In one embodiment of the present invention, by The fused image is subjected to noise removal processing to obtain a first feature image, wherein is the standard deviation of the Gaussian filter, represents the weight coefficient of the Gaussian filter, Indicates that it is located The Gaussian function value of is the weight of the point.

[0020] In one embodiment of the present invention, characters in each text block are recognized to obtain a text string, and the recognized text string is semantically parsed using natural language processing technology to extract field content and meaning, and the recognized field content is matched with the rules according to a predefined rule base to determine the category of the field, wherein the rule base contains field names and variants of field names, for example: Amount: ["amount", "total price", "total", "total amount"], Date: ["Date", "Time", "Issuance Date", "Transaction Date"], InvoiceNumber: ["invoiceNumber","invoiceNumber","number","invoiceNumber"].

[0021] In one embodiment of the present invention, the semantic analysis of the identified text string using natural language processing technology includes: Step S501, segmenting a continuous text string, for example, segmenting the string "total amount: 100 yuan" into "total amount", ":", "100", "yuan"; Step S502, perform part-of-speech tagging on the segmented words, for example, "total amount" is marked as a noun, "100" is marked as a numeral, and "yuan" is marked as a quantifier; Step S503, performing entity recognition on the words after part-of-speech tagging, for example, identifying "100 yuan" as a money entity.

[0022] In one embodiment of the present invention, a transfer-based dependency parsing algorithm is used to analyze the logical relationship between fields in receipt data. The specific steps include: first, converting the field content into an input format suitable for dependency parsing, then building a dependency tree by applying the LEFT-ARC transfer operation to form a dependency graph between fields, and then identifying the logical relationship between fields. Finally, matching the field content is completed according to the dependency graph.

[0023] In one embodiment of the present invention, the matched receipt data is cleaned and content verified. Data cleaning includes: formatting the field content, removing spaces and redundant characters, and spelling correction. Content verification includes: format verification, logical relationship verification and consistency check of the field; for example, date format verification is performed on the field to ensure that the date field conforms to the "YYYY-MM-DD" format; the numeric format verification is performed on the field to ensure that the numeric field is a valid number; the amount calculation verification is performed to ensure that the "amount" is equal to the "unit price" multiplied by the "quantity"; the invoice number consistency verification is performed on the field to ensure that the "invoice number" is consistent in all related fields.

[0024] In one embodiment of the present invention, the cleaned and content-verified receipt data is converted into a standardized structured output to ensure that the data format is unified and easy to read. The data output formats include: JSON, XML, CSV and database table structure.

[0025] In one embodiment of the present invention, a method for identifying clearing receipt data is also provided, comprising the following steps: Step S601, taking five images of the target receipt from multiple angles, aligning them using image registration technology, fusing the aligned images using a fusion algorithm to obtain a fused image, and performing noise reduction and contrast enhancement on the fused image to obtain a first feature image; Step S602, extracting receipt data from the first feature image using image recognition technology; Step S603, identifying the content and meaning of the fields in the receipt data, and classifying the fields; Step S604, using a dependency parsing algorithm to identify the logical relationship between the fields in the receipt data, and matching the fields in the receipt data; Step S605, cleaning and content verification of the matched receipt data; Step S606: Output the cleaned and verified receipt data in a structured manner.

[0026] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A clearing receipt data recognition system, characterized in that: It includes image acquisition and preprocessing module, image recognition module, field recognition and classification module, field association analysis module, data cleaning and verification module and data output module: The image acquisition and preprocessing module is used to take five images of the target receipt from multiple angles and align them using image registration technology, specifically as follows: take images from the front, upper left, lower left, upper right and lower right angles of the target receipt, identify feature points in each image using a scale-invariant feature transformation algorithm, output the coordinates and feature descriptors of the feature points, calculate the similarity between the feature points using Euclidean distance, match the feature points, eliminate mismatches by calculating the distance ratio, and align similar areas of the five images using affine transformation; The steps of using the scale-invariant feature transformation algorithm include: Step S201, generating Gaussian difference images of different scales based on five images at different angles; Step S202, finding potential feature points from the Gaussian difference image by extreme value detection, filtering edge responses and removing low contrast of the potential feature points to obtain feature points, and extracting the coordinates of the feature points; Step S203, assigning a main direction to each feature point by calculating the gradient direction histogram of the feature point neighborhood; Step S204, concatenating the gradient direction histograms of the feature point neighborhood to obtain a feature descriptor of the feature point; The distance ratio is calculated as follows: ,in, represents the distance ratio, Indicates the distance between the current feature point descriptor and the nearest feature point. Indicates the distance between the current feature point descriptor and the next closest feature point; when the distance ratio is greater than the third threshold, the match is a false match and needs to be removed. The third threshold is a custom parameter; The step of aligning similar regions of the five images using affine transformation comprises: Step S301, the image taken at the front angle is used as the reference image, and another image is selected at the same time, K pairs of matching feature points are found in the two images, and an affine transformation matrix is ​​constructed based on the coordinate transformation formula of the affine transformation principle and the matching feature points. The matrix expression is: , where T represents the affine transformation matrix, a, b, c and d represent the first coefficient, the second coefficient, the third coefficient and the fourth coefficient respectively, which are used to control the scaling, rotation and tilt of the image, tx and ty represent the first translation parameter and the second translation parameter respectively, and the coordinate transformation formula of the affine transformation principle is: ,in and are the horizontal and vertical coordinates of the pixel coordinates after transformation, respectively. x and y are the horizontal and vertical coordinates of the pixel coordinates before transformation, respectively. a, b, c, d, tx and ty are solved by solving a linear equation system constructed by matching feature points and combining coordinate transformation formula. Step S302, using a random sampling consistency algorithm to filter out mismatched feature points; Step S303, using an affine transformation matrix, according to the coordinate transformation formula in step S301, the pixel coordinates of another image are calculated and transformed through the pixel coordinates of the reference image; Step S304, repeating steps S301 to S303, applying the transformation to all images except the reference image to align them with the reference image; Applying a fusion algorithm to fuse the aligned images to obtain a fused image, performing noise reduction and contrast enhancement processing on the fused image to obtain a first feature image; An image recognition module, configured to extract receipt data from the first feature image using image recognition technology; The field recognition and classification module is used to recognize the content and meaning of the fields in the receipt data and classify the fields, including: extracting all text blocks from the first feature image using optical character recognition technology to obtain receipt data, wherein each text block contains a group of continuous characters; A field association analysis module, used to identify the logical relationship between the fields in the receipt data using a dependency syntax analysis algorithm, and match the fields in the receipt data; Data cleaning and verification module, used to clean and verify the content of the matched receipt data; The data output module is used to output the receipt data after cleaning and verification in a structured manner.

2. A liquidation receipt data recognition system according to claim 1, characterized in that: Use the fusion algorithm to fuse the aligned images, including: Step S401, defining a neighborhood window of size N×N around each pixel of each image, counting the frequency distribution of each gray level in the neighborhood window, and generating a gray level histogram of the neighborhood window, wherein N is a custom parameter and the gray level represents the brightness of the pixel in the image; Step S402, calculating the entropy value of the pixel point according to the grayscale histogram, and selecting the pixel with the highest entropy value in the same pixel position in the five images. The entropy value calculation formula is: , where H represents the entropy value of the pixel, represents the frequency of gray level i appearing in the neighborhood window, and J represents the number of gray levels; Step S403, filling the pixel with the maximum entropy value at each pixel position into the corresponding position of the final fused image to generate a complete fused image.

3. A liquidation receipt data recognition system according to claim 1, characterized in that: The characters in each text block are recognized to obtain a text string, and the recognized text string is semantically parsed using natural language processing technology to extract field content and meaning. The recognized field content is matched with the rules according to a predefined rule base to determine the category of the field, wherein the rule base contains field names and variants of field names.

4. A liquidation receipt data recognition system according to claim 1, characterized in that: A transfer-based dependency parsing algorithm is used to analyze the logical relationship between fields in receipt data. The specific steps include: first, converting the field content into an input format suitable for dependency parsing, then building a dependency tree by applying the LEFT-ARC transfer operation to form a dependency graph between fields, and then identifying the logical relationship between fields. Finally, matching the field content is completed according to the dependency graph.

5. A liquidation receipt data recognition system according to claim 1, characterized in that: The matched receipt data is cleaned and content verified. Data cleaning includes: formatting the field content, removing spaces and redundant characters, and spelling correction; content verification includes: field format verification, logical relationship verification and consistency check.

6. A method for identifying liquidation receipt data, characterized in that: A clearing receipt data recognition system applied to any one of claims 1 to 5, comprising the following steps: Step S601, taking five images of the target receipt from multiple angles, aligning them using image registration technology, fusing the aligned images using a fusion algorithm to obtain a fused image, and performing noise reduction and contrast enhancement on the fused image to obtain a first feature image; Step S602, extracting receipt data from the first feature image using image recognition technology; Step S603, identifying the content and meaning of the fields in the receipt data, and classifying the fields; Step S604, using a dependency parsing algorithm to identify the logical relationship between the fields in the receipt data, and matching the fields in the receipt data; Step S605, cleaning and content verification of the matched receipt data; Step S606: Output the cleaned and verified receipt data in a structured manner.

Images