Jewelry information identification and proofreading method and system based on deep learning

Abstract

The invention discloses a deep learning-based jewelry information identification and proofreading method and system, and the method comprises the following steps: collecting a jewelry authentication certificate image and a jewelry label image, and carrying out the preprocessing of the images; respectively identifying character information of data areas of interested areas in the preprocessed jewelry identification certificate image and the jewelry label image through an OCR (Optical Character Recognition) technology; respectively carrying out data processing on character information of data areas of interested areas in the preprocessed jewelry authentication certificate image and the jewelry label image to obtain jewelry information of a jewelry authentication certificate and jewelry information of a jewelry label; and comparing the jewelry information of the jewelry authentication certificate with the jewelry information of the jewelry label. According to the invention, the problems of low checking efficiency and increased error rate caused by the fact that the quality of the jewelry authentication certificate is checked manually in the prior art, which is time-consuming and labor-consuming, and is easily influenced by human factors are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of jewelry information processing, and in particular to a jewelry information recognition and proofreading method and system based on deep learning. Background Art

[0002] In the production and sales process of the jewelry industry, jewelry appraisal certificates are important legal documents that confirm the authenticity, quality and value of jewelry. The accuracy of the information on the jewelry appraisal certificate plays an irreplaceable role in safeguarding consumer rights and ensuring market integrity. Traditionally, this key link relies on manual operation, that is, the staff needs to compare and verify the information on the jewelry appraisal certificate with the information on the tag attached to the actual jewelry one by one. This manual verification method is not only time-consuming and labor-intensive, but also easily affected by human factors, such as visual fatigue and distraction, which leads to low verification efficiency and increased error rate. Summary of the invention

[0003] In response to the above-mentioned defects, the present invention proposes a jewelry information recognition and proofreading method and system based on deep learning, aiming to solve the problem of traditional manual verification of the quality of jewelry appraisal certificates, which is not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in low verification efficiency and increased error rate.

[0004] To achieve this object, the present invention adopts the following technical solutions:

[0005] A jewelry information recognition and proofreading method based on deep learning, comprising the following steps:

[0006] Step S1: Collecting jewelry appraisal certificate images and jewelry label images;

[0007] Step S2: performing image preprocessing on the jewelry appraisal certificate image and the jewelry label image respectively to obtain a preprocessed jewelry appraisal certificate image and a preprocessed jewelry label image;

[0008] Step S3: extracting regions of interest from the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image respectively, wherein the regions of interest include a plurality of header regions and data regions, and each header region corresponds to one or more data regions;

[0009] Step S4: locating the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and locating the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features;

[0010] Step S5: determining the corresponding data region according to the header region of the region of interest in the preprocessed jewelry appraisal certificate image; and determining the corresponding data region according to the header region of the region of interest in the preprocessed jewelry label image;

[0011] Step S6: identifying text information of the data area of ​​the area of ​​interest in the pre-processed jewelry appraisal certificate image by using the OCR technology; and identifying text information of the data area of ​​the area of ​​interest in the pre-processed jewelry label image by using the OCR technology;

[0012] Step S7: performing data processing on the text information in the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain jewelry information of the jewelry appraisal certificate; and performing data processing on the text information in the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain jewelry information of the jewelry label;

[0013] Step S8: Compare the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0014] Preferably, in step S2, image preprocessing is performed on the jewelry appraisal certificate image and the jewelry label image respectively, which specifically includes the following sub-steps:

[0015] The jewelry appraisal certificate image is processed by grayscale stretching and high-frequency filtering, and the jewelry label image is processed by grayscale stretching and high-frequency filtering.

[0016] Preferably, in step S3, in the process of extracting the region of interest in the preprocessed jewelry label image, the following steps are also included:

[0017] The bottom text area of ​​the area of ​​interest in the preprocessed jewelry label image is extracted through threshold segmentation, connected components and horizontal closing operations;

[0018] Convert the bottom text area into a minimum bounding rectangle;

[0019] Along the downward direction of the minimum circumscribed rectangle, search for a gray value jump point every 10 pixels to obtain a number of gray value jump points;

[0020] Fitting several grayscale value jump points into straight lines to obtain a reference line for the jewelry label;

[0021] Calculate the distances from the lower left corner and the lower right corner of the minimum circumscribed rectangle to the reference line of the jewelry label, and take the minimum value of the distances as the distance from the bottom text area to the reference line of the jewelry label;

[0022] It is determined whether the distance between the bottom text area and the reference line of the jewelry label is within 2-20 pixels. If so, it means that the position of the bottom text area in the area of ​​interest is qualified. If not, it means that the position of the bottom text area in the area of ​​interest is unqualified.

[0023] Preferably, in step S7, data processing is performed on the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image, specifically including the following sub-steps: filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image;

[0024] Data processing is performed on the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image, which specifically includes the following sub-steps: filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image.

[0025] Preferably, in step S8, in addition to comparing the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label, the following steps are also included:

[0026] Obtain the jewelry quality in the jewelry information on the jewelry appraisal certificate, the jewelry quality in the jewelry information on the jewelry label, and the actual quality of the jewelry;

[0027] Determine whether the jewelry quality in the jewelry information on the jewelry appraisal certificate, the jewelry quality in the jewelry information on the jewelry label, and the actual quality of the jewelry meet the following conditions:

[0028] |M1-M3|≤d and |M2-M3|≤d;

[0029] Among them, M1 represents the jewelry quality in the jewelry information of the jewelry appraisal certificate; M2 represents the jewelry quality in the jewelry information of the jewelry label; M3 represents the actual quality of the jewelry; d represents the allowable quality tolerance;

[0030] If it is met, it proves that the quality of the jewelry appraisal certificate is qualified; if it is not met, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0031] Another aspect of the present application provides a jewelry information recognition and proofreading system based on deep learning, the system comprising:

[0032] A collection module, used for collecting jewelry appraisal certificate images and jewelry label images;

[0033] A preprocessing module, used to perform image preprocessing on the jewelry appraisal certificate image and the jewelry label image respectively, to obtain a preprocessed jewelry appraisal certificate image and a preprocessed jewelry label image;

[0034] A first extraction module is used to extract regions of interest from the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image, respectively, wherein the regions of interest include a plurality of header regions and data regions, and each header region corresponds to one or more data regions;

[0035] A positioning module, for positioning the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and positioning the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features;

[0036] A determination module, for determining a corresponding data region according to a header region of an area of ​​interest in a preprocessed jewelry appraisal certificate image; and a corresponding data region according to a header region of an area of ​​interest in a preprocessed jewelry label image;

[0037] A recognition module, used to recognize text information of a data area of ​​an area of ​​interest in a pre-processed jewelry appraisal certificate image by using an OCR technology; and recognize text information of a data area of ​​an area of ​​interest in a pre-processed jewelry label image by using an OCR technology;

[0038] A data processing module is used to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain the jewelry information of the jewelry appraisal certificate; and to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain the jewelry information of the jewelry label;

[0039] The comparison module is used to compare the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0040] Preferably, the preprocessing module comprises:

[0041] The first data processing submodule is used to perform grayscale stretching and high-frequency filtering on the jewelry appraisal certificate image;

[0042] The second data processing submodule is used to perform grayscale stretching and high-frequency filtering on the jewelry label image.

[0043] Preferably, during the execution of the first extraction module, the process further includes:

[0044] A second extraction module is used to extract the bottom text area of ​​the area of ​​interest in the pre-processed jewelry label image through threshold segmentation, connected components and horizontal closing operations;

[0045] A conversion module, used for converting the bottom text area into a minimum bounding rectangle;

[0046] A search module is used to search for a gray value jump point every 10 pixels along the downward direction of the minimum circumscribed rectangle to obtain a plurality of gray value jump points;

[0047] A calculation module, used for calculating the distances from the lower left corner and the lower right corner of the minimum circumscribed rectangle to the reference line of the jewelry label, and taking the minimum value of the distances as the distance from the bottom text area to the reference line of the jewelry label;

[0048] The first judgment module is used to judge whether the distance between the bottom text area and the reference line of the jewelry label is within 2-20 pixels. If so, it means that the position of the bottom text area in the area of ​​interest is qualified. If not, it means that the position of the bottom text area in the area of ​​interest is unqualified.

[0049] Preferably, the data processing module includes:

[0050] A first filtering submodule is used to filter the text information in the data area of ​​the area of ​​interest in the pre-processed jewelry appraisal certificate image;

[0051] The second filtering submodule is used to filter the text information of the data area of ​​the area of ​​interest in the pre-processed jewelry label image.

[0052] Preferably, it also includes:

[0053] An acquisition module, used to acquire the jewelry quality in the jewelry information of the jewelry appraisal certificate, the jewelry quality in the jewelry information of the jewelry label, and the actual quality of the jewelry;

[0054] The second judgment module is used to judge whether the jewelry quality in the jewelry information of the jewelry appraisal certificate, the jewelry quality in the jewelry information of the jewelry label and the actual quality of the jewelry meet the following conditions:

[0055] |M1-M3|≤d and |M2-M3|≤d;

[0056] Among them, M1 represents the jewelry quality in the jewelry information of the jewelry appraisal certificate; M2 represents the jewelry quality in the jewelry information of the jewelry label; M3 represents the actual quality of the jewelry; d represents the allowable quality tolerance;

[0057] If it is met, it proves that the quality of the jewelry appraisal certificate is qualified; if it is not met, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0058] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0059] In this solution, by collecting and processing jewelry appraisal certificate images and jewelry label images, and using OCR technology to identify the jewelry information in the processed jewelry appraisal certificate images and jewelry label images, and by judging whether the jewelry information in the processed jewelry appraisal certificate is consistent with the jewelry information in the processed jewelry label, the quality of the jewelry appraisal certificate is automatically tested. Compared with the traditional manual method of testing the quality of jewelry appraisal certificates, this solution can effectively save manpower, improve verification efficiency and reduce the error rate. It is further explained that this automated detection is compatible with different types of jewelry appraisal certificates, such as NGTC, IGI and GIA certificates. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flowchart of the steps of jewelry information identification and proofreading method based on deep learning. DETAILED DESCRIPTION

[0061] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0062] A jewelry information recognition and proofreading method based on deep learning, comprising the following steps:

[0063] Step S1: Collecting jewelry appraisal certificate images and jewelry label images;

[0064] Step S2: performing image preprocessing on the jewelry appraisal certificate image and the jewelry label image respectively to obtain a preprocessed jewelry appraisal certificate image and a preprocessed jewelry label image;

[0065] Step S3: extracting regions of interest from the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image respectively, wherein the regions of interest include a plurality of header regions and data regions, and each header region corresponds to one or more data regions;

[0066] Step S4: locating the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and locating the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features;

[0067] Step S5: determining the corresponding data region according to the header region of the region of interest in the preprocessed jewelry appraisal certificate image; and determining the corresponding data region according to the header region of the region of interest in the preprocessed jewelry label image;

[0068] Step S6: identifying text information of the data area of ​​the area of ​​interest in the pre-processed jewelry appraisal certificate image by using the OCR technology; and identifying text information of the data area of ​​the area of ​​interest in the pre-processed jewelry label image by using the OCR technology;

[0069] Step S7: performing data processing on the text information in the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain jewelry information of the jewelry appraisal certificate; and performing data processing on the text information in the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain jewelry information of the jewelry label;

[0070] Step S8: Compare the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0071] This scheme is a jewelry information recognition and proofreading method based on deep learning, such as Figure 1As shown, the first step is to collect the jewelry appraisal certificate image and the jewelry label image. Specifically, the image of the jewelry appraisal certificate and the image of the tag label attached to the actual jewelry are collected by a camera. The second step is to preprocess the jewelry appraisal certificate image and the jewelry label image respectively to obtain the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image. In this embodiment, by preprocessing the jewelry appraisal certificate image and the jewelry label image, the interference of the background in the jewelry appraisal certificate image and the jewelry label image can be reduced, and the text information in the jewelry appraisal certificate image and the jewelry label image can be highlighted. The third step is to extract the regions of interest in the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image respectively, wherein the regions of interest include several header areas and data areas, and each header area corresponds to one or more data areas. In this embodiment, by extracting the regions of interest, the required information in the image can be quickly extracted, thereby reducing unnecessary data processing. The fourth step is to locate the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and locate the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features. In this embodiment, the header area has characteristic information of a specific geometric structure, and the header area in the area of ​​interest can be quickly located by the characteristic information of this specific geometric structure. The fifth step is to determine the corresponding data area according to the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image; and determine the corresponding data area of ​​the header area of ​​the area of ​​interest in the preprocessed jewelry label image. In this embodiment, in the area of ​​interest, the position of the data area and the position of the header area are on the same horizontal line, so the position of the data area can be horizontally located by the position of the header area. The sixth step is to identify the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by OCR technology; and identify the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image by OCR technology. Specifically, OCR technology is an existing text recognition technology, and OCR technology can quickly identify the text information in the data area. In one embodiment, taking the jewelry appraisal certificate for 18K gold diamond earrings with a total mass of 3.918g, a round diamond shape, a colorless color, and a feather texture as an example, the OCR technology is used to identify that the data text information corresponding to the header "Test Conclusion" is "18K Gold Diamond Earrings", the data text information corresponding to the header "Total Mass" is "3.918g", the data text information corresponding to the header "Shape" is "round diamond shape", the data text information corresponding to the header "Color" is "colorless", the data text information corresponding to the header "Magnified Inspection" is "Feathered", and the data text information corresponding to the header "Precious Metal Detection" is "18K", etc.The seventh step is to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain the jewelry information of the jewelry appraisal certificate; and to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain the jewelry information of the jewelry label. Specifically, the text information obtained by OCR technology recognition may be interfered by noise, so that the text information contains some redundant information. Therefore, it is necessary to process the text information obtained by OCR technology recognition to remove the redundant information and ensure the quality of the text information. The eighth step is to compare the jewelry information of the jewelry appraisal certificate with the jewelry information of the jewelry label to determine whether the jewelry information of the jewelry appraisal certificate is consistent with the jewelry information of the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified. In this embodiment, by determining whether the jewelry information of the jewelry appraisal certificate is consistent with the jewelry information of the jewelry label, it is possible to effectively determine whether the quality of the jewelry appraisal certificate is qualified. When an unqualified jewelry appraisal certificate is found, the content of the certificate can be changed in time to prevent the problematic certificate from flowing out.

[0072] In this solution, by collecting and processing jewelry appraisal certificate images and jewelry label images, and using OCR technology to identify the jewelry information in the processed jewelry appraisal certificate images and jewelry label images, and by judging whether the jewelry information in the processed jewelry appraisal certificate is consistent with the jewelry information in the processed jewelry label, the quality of the jewelry appraisal certificate is automatically tested. Compared with the traditional manual method of testing the quality of jewelry appraisal certificates, this solution can effectively save manpower, improve verification efficiency and reduce the error rate. It is further explained that this automated detection is compatible with different types of jewelry appraisal certificates, such as NGTC, IGI and GIA certificates.

[0073] Preferably, in step S2, the jewelry appraisal certificate image and the jewelry label image are respectively subjected to image preprocessing, specifically including the following sub-steps: grayscale stretching and high-frequency filtering processing on the jewelry appraisal certificate image, and grayscale stretching and high-frequency filtering processing on the jewelry label image.

[0074] In this embodiment, by performing grayscale stretching processing on the jewelry appraisal certificate image and the jewelry label image, the contrast of the jewelry appraisal certificate image and the jewelry label image can be enhanced. By performing high-frequency filtering processing on the jewelry appraisal certificate image and the jewelry label image, the edge details in the jewelry appraisal certificate image and the jewelry label image can be enhanced, thereby highlighting the text display.

[0075] Preferably, in step S3, in the process of extracting the region of interest in the preprocessed jewelry label image, the following steps are also included:

[0076] The bottom text area of ​​the area of ​​interest in the preprocessed jewelry label image is extracted through threshold segmentation, connected components and horizontal closing operations;

[0077] Convert the bottom text area into a minimum bounding rectangle;

[0078] Along the downward direction of the minimum circumscribed rectangle, search for a gray value jump point every 10 pixels to obtain a number of gray value jump points;

[0079] Fitting several grayscale value jump points into straight lines to obtain a reference line for the jewelry label;

[0080] Calculate the distances from the lower left corner and the lower right corner of the minimum circumscribed rectangle to the reference line of the jewelry label, and take the minimum value of the distances as the distance from the bottom text area to the reference line of the jewelry label;

[0081] It is determined whether the distance between the bottom text area and the reference line of the jewelry label is within 2-20 pixels. If so, it means that the position of the bottom text area in the area of ​​interest is qualified. If not, it means that the position of the bottom text area in the area of ​​interest is unqualified.

[0082] In this embodiment, in the process of extracting the region of interest in the preprocessed jewelry label image, it can be determined whether the printing position of the bottom text in the jewelry label is correct, that is, whether the distance from the bottom text area to the reference line of the jewelry label is within 2-20 pixels. If the distance from the bottom text area to the reference line of the jewelry label is less than 2 pixels, it is considered that the distance from the bottom text area to the reference line of the jewelry label is too close. If the distance from the bottom text area to the reference line of the jewelry label is greater than 20 pixels, it is considered that the distance from the bottom text area to the reference line of the jewelry label is too far. Further explanation, threshold segmentation, connected components and horizontal closing operations are all prior arts. In the process of converting the bottom text area into a minimum bounding rectangle, the calculation principle of the minimum bounding rectangle is: determine a minimum rectangle that contains all pixels in a given area. The gray value jump point represents the point of transition from black to white.

[0083] Preferably, in step S7, data processing is performed on the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image, specifically including the following sub-steps: filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image;

[0084] Data processing is performed on the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image, which specifically includes the following sub-steps: filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image.

[0085] In this embodiment, by filtering the preprocessed jewelry appraisal certificate image and the text information in the data area of ​​the area of ​​interest in the jewelry label, specifically filtering, the noise signal in the text information can be removed to improve the clarity and readability of the text information.

[0086] Preferably, in step S8, in addition to comparing the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label, the following steps are also included:

[0087] Obtain the jewelry quality in the jewelry information on the jewelry appraisal certificate, the jewelry quality in the jewelry information on the jewelry label, and the actual quality of the jewelry;

[0088] Determine whether the jewelry quality in the jewelry information on the jewelry appraisal certificate, the jewelry quality in the jewelry information on the jewelry label, and the actual quality of the jewelry meet the following conditions:

[0089] |M1-M3|≤d and |M2-M3|≤d;

[0090] Among them, M1 represents the jewelry quality in the jewelry information of the jewelry appraisal certificate; M2 represents the jewelry quality in the jewelry information of the jewelry label; M3 represents the actual quality of the jewelry; d represents the allowable quality tolerance;

[0091] If it is met, it proves that the quality of the jewelry appraisal certificate is qualified; if it is not met, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0092] In this embodiment, the jewelry is weighed by an electronic scale to obtain the actual mass of the jewelry, and the jewelry information in the jewelry appraisal certificate and the jewelry label identified by OCR technology is used to check whether the actual mass of the jewelry, the jewelry mass in the jewelry appraisal certificate, and the jewelry mass in the jewelry label are within the allowable quality tolerance range to further determine whether the quality of the jewelry appraisal certificate is qualified.

[0093] When d is 0.01g, M1 is 3.91g, M2 is 3.91g, and M3 is 3.90g, then |M1-M3|=0.01g≤0.01g and |M2-M3|=0.01g≤0.01g. It is considered that the actual quality of the jewelry, the jewelry quality in the jewelry information on the jewelry appraisal certificate, and the jewelry quality in the jewelry information on the jewelry label are within the allowable quality tolerance range, further indicating that the quality of the jewelry appraisal certificate is qualified.

[0094] Another aspect of the present application provides a jewelry information recognition and proofreading system based on deep learning, the system comprising:

[0095] A collection module, used for collecting jewelry appraisal certificate images and jewelry label images;

[0096] A preprocessing module, used to perform image preprocessing on the jewelry appraisal certificate image and the jewelry label image respectively, to obtain a preprocessed jewelry appraisal certificate image and a preprocessed jewelry label image;

[0097] A first extraction module is used to extract regions of interest from the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image, respectively, wherein the regions of interest include a plurality of header regions and data regions, and each header region corresponds to one or more data regions;

[0098] A positioning module, for positioning the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and positioning the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features;

[0099] A determination module, for determining a corresponding data region according to a header region of an area of ​​interest in a preprocessed jewelry appraisal certificate image; and a corresponding data region according to a header region of an area of ​​interest in a preprocessed jewelry label image;

[0100] A recognition module, used to recognize text information of a data area of ​​an area of ​​interest in a pre-processed jewelry appraisal certificate image by using an OCR technology; and recognize text information of a data area of ​​an area of ​​interest in a pre-processed jewelry label image by using an OCR technology;

[0101] A data processing module is used to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain the jewelry information of the jewelry appraisal certificate; and to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain the jewelry information of the jewelry label;

[0102] The comparison module is used to compare the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0103] This scheme is a jewelry information recognition and proofreading system based on deep learning, which realizes the automatic detection of the quality of jewelry appraisal certificates through the cooperation of the acquisition module, preprocessing module, first extraction module, positioning module, determination module, recognition module, data processing module and comparison module. Compared with the traditional manual method of detecting the quality of jewelry appraisal certificates, this scheme can effectively save manpower, improve verification efficiency and reduce error rate.

[0104] Preferably, the preprocessing module includes: a first data processing submodule for grayscale stretching and high-frequency filtering of the jewelry appraisal certificate image; and a second data processing submodule for grayscale stretching and high-frequency filtering of the jewelry label image. In this embodiment, by grayscale stretching of the jewelry appraisal certificate image and the jewelry label image, the contrast of the jewelry appraisal certificate image and the jewelry label image can be enhanced. By high-frequency filtering of the jewelry appraisal certificate image and the jewelry label image, the edge details in the jewelry appraisal certificate image and the jewelry label image can be enhanced, thereby highlighting the text display.

[0105] Preferably, during the execution of the first extraction module, the process further includes:

[0106] A second extraction module is used to extract the bottom text area of ​​the area of ​​interest in the pre-processed jewelry label image through threshold segmentation, connected components and horizontal closing operations;

[0107] A conversion module, used for converting the bottom text area into a minimum bounding rectangle;

[0108] A search module is used to search for a gray value jump point every 10 pixels along the downward direction of the minimum circumscribed rectangle to obtain a plurality of gray value jump points;

[0109] A calculation module, used for calculating the distances from the lower left corner and the lower right corner of the minimum circumscribed rectangle to the reference line of the jewelry label, and taking the minimum value of the distances as the distance from the bottom text area to the reference line of the jewelry label;

[0110] The first judgment module is used to judge whether the distance between the bottom text area and the reference line of the jewelry label is within 2-20 pixels. If so, it means that the position of the bottom text area in the area of ​​interest is qualified. If not, it means that the position of the bottom text area in the area of ​​interest is unqualified.

[0111] In this embodiment, through the cooperation of the second extraction module, the conversion module, the search module, the calculation module and the first judgment module, it is possible to identify whether the printing position of the bottom text in the jewelry label is correct.

[0112] Preferably, the data processing module includes: a first filtering submodule for filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image; a second filtering submodule for filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image. In this embodiment, by setting the first filtering submodule and the second filtering submodule, the noise signal in the text information can be removed, and the clarity and readability of the text information can be improved.

[0113] Preferably, it also includes:

[0114] An acquisition module, used to acquire the jewelry quality in the jewelry information of the jewelry appraisal certificate, the jewelry quality in the jewelry information of the jewelry label, and the actual quality of the jewelry;

[0115] The second judgment module is used to judge whether the jewelry quality in the jewelry information of the jewelry appraisal certificate, the jewelry quality in the jewelry information of the jewelry label and the actual quality of the jewelry meet the following conditions:

[0116] |M1-M3|≤d and |M2-M3|≤d;

[0117] Among them, M1 represents the jewelry quality in the jewelry information of the jewelry appraisal certificate; M2 represents the jewelry quality in the jewelry information of the jewelry label; M3 represents the actual quality of the jewelry; d represents the allowable quality tolerance; if it is met, it proves that the quality of the jewelry appraisal certificate is qualified; if it is not met, it proves that the quality of the jewelry appraisal certificate is unqualified.

[0118] In this embodiment, in the acquisition module, the jewelry is weighed by an electronic scale device to obtain the actual mass of the jewelry, and the jewelry information in the jewelry appraisal certificate and the jewelry label is identified by OCR technology. In the second judgment module, by checking whether the actual mass of the jewelry, the jewelry mass in the jewelry appraisal certificate, and the jewelry mass in the jewelry label are within the allowable quality tolerance range, it can be further judged whether the quality of the jewelry appraisal certificate is qualified.

[0119] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0120] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A jewelry information recognition and proofreading method based on deep learning, characterized in that: The following steps are involved: Step S1: Collecting jewelry appraisal certificate images and jewelry label images; Step S2: performing image preprocessing on the jewelry appraisal certificate image and the jewelry label image respectively to obtain a preprocessed jewelry appraisal certificate image and a preprocessed jewelry label image; Step S3: extracting regions of interest from the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image respectively, wherein the regions of interest include a plurality of header regions and data regions, and each header region corresponds to one or more data regions; Step S4: locating the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and locating the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features; Step S5: determining the corresponding data region according to the header region of the region of interest in the preprocessed jewelry appraisal certificate image; and determining the corresponding data region according to the header region of the region of interest in the preprocessed jewelry label image; Step S6: identifying text information of the data area of ​​the area of ​​interest in the pre-processed jewelry appraisal certificate image by using the OCR technology; and identifying text information of the data area of ​​the area of ​​interest in the pre-processed jewelry label image by using the OCR technology; Step S7: performing data processing on the text information in the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain jewelry information of the jewelry appraisal certificate; and performing data processing on the text information in the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain jewelry information of the jewelry label; Step S8: Compare the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified.

2. The jewelry information recognition and proofreading method based on deep learning according to claim 1, characterized in that: In step S2, the jewelry appraisal certificate image and the jewelry label image are respectively preprocessed, which specifically includes the following sub-steps: The jewelry appraisal certificate image is processed by grayscale stretching and high-frequency filtering, and the jewelry label image is processed by grayscale stretching and high-frequency filtering.

3. The jewelry information recognition and proofreading method based on deep learning according to claim 1, characterized in that: In step S3, the process of extracting the region of interest in the preprocessed jewelry label image further includes the following steps: The bottom text area of ​​the area of ​​interest in the preprocessed jewelry label image is extracted through threshold segmentation, connected components and horizontal closing operations; Convert the bottom text area into a minimum bounding rectangle; Along the downward direction of the minimum circumscribed rectangle, search for a gray value jump point every 10 pixels to obtain a number of gray value jump points; Fitting several grayscale value jump points into straight lines to obtain a reference line for the jewelry label; Calculate the distances from the lower left corner and the lower right corner of the minimum circumscribed rectangle to the reference line of the jewelry label, and take the minimum value of the distances as the distance from the bottom text area to the reference line of the jewelry label; It is determined whether the distance between the bottom text area and the reference line of the jewelry label is within 2-20 pixels. If so, it means that the position of the bottom text area in the area of ​​interest is qualified. If not, it means that the position of the bottom text area in the area of ​​interest is unqualified.

4. The jewelry information recognition and proofreading method based on deep learning according to claim 1, characterized in that: In step S7, data processing is performed on the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image, which specifically includes the following sub-steps: filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image; Data processing is performed on the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image, which specifically includes the following sub-steps: filtering the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image.

5. The jewelry information recognition and proofreading method based on deep learning according to claim 1, characterized in that: In step S8, in addition to comparing the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label, the following steps are also included: Obtain the jewelry quality in the jewelry information on the jewelry appraisal certificate, the jewelry quality in the jewelry information on the jewelry label, and the actual quality of the jewelry; Determine whether the jewelry quality in the jewelry information on the jewelry appraisal certificate, the jewelry quality in the jewelry information on the jewelry label, and the actual quality of the jewelry meet the following conditions: |M1-M3|≤d and |M2-M3|≤d; Among them, M1 represents the jewelry quality in the jewelry information of the jewelry appraisal certificate; M2 represents the jewelry quality in the jewelry information of the jewelry label; M3 represents the actual quality of the jewelry; d represents the allowable quality tolerance; If it is met, it proves that the quality of the jewelry appraisal certificate is qualified; if it is not met, it proves that the quality of the jewelry appraisal certificate is unqualified.

6. A jewelry information recognition and proofreading system based on deep learning, using the jewelry information recognition and proofreading method based on deep learning as claimed in any one of claims 1 to 5, characterized in that: The system comprises: A collection module, used for collecting jewelry appraisal certificate images and jewelry label images; A preprocessing module, used to perform image preprocessing on the jewelry appraisal certificate image and the jewelry label image respectively, to obtain a preprocessed jewelry appraisal certificate image and a preprocessed jewelry label image; A first extraction module is used to extract regions of interest from the preprocessed jewelry appraisal certificate image and the preprocessed jewelry label image, respectively, wherein the regions of interest include a plurality of header regions and data regions, and each header region corresponds to one or more data regions; A positioning module, for positioning the header area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image by shape features; and positioning the header area of ​​the area of ​​interest in the preprocessed jewelry label image by shape features; A determination module, for determining a corresponding data region according to a header region of an area of ​​interest in a preprocessed jewelry appraisal certificate image; and a corresponding data region according to a header region of an area of ​​interest in a preprocessed jewelry label image; A recognition module, used to recognize text information of a data area of ​​an area of ​​interest in a pre-processed jewelry appraisal certificate image by using an OCR technology; and recognize text information of a data area of ​​an area of ​​interest in a pre-processed jewelry label image by using an OCR technology; A data processing module is used to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry appraisal certificate image to obtain the jewelry information of the jewelry appraisal certificate; and to process the text information of the data area of ​​the area of ​​interest in the preprocessed jewelry label image to obtain the jewelry information of the jewelry label; The comparison module is used to compare the jewelry information on the jewelry appraisal certificate with the jewelry information on the jewelry label to determine whether the jewelry information on the jewelry appraisal certificate is consistent with the jewelry information on the jewelry label. If they are consistent, it proves that the quality of the jewelry appraisal certificate is qualified; if they are inconsistent, it proves that the quality of the jewelry appraisal certificate is unqualified.

7. The jewelry information recognition and proofreading system based on deep learning according to claim 6, characterized in that: The preprocessing module comprises: The first data processing submodule is used to perform grayscale stretching and high-frequency filtering on the jewelry appraisal certificate image; The second data processing submodule is used to perform grayscale stretching and high-frequency filtering on the jewelry label image.

8. The jewelry information recognition and proofreading system based on deep learning according to claim 6, characterized in that: In the process of executing the first extraction module, it also includes: A second extraction module is used to extract the bottom text area of ​​the area of ​​interest in the pre-processed jewelry label image through threshold segmentation, connected components and horizontal closing operations; A conversion module, used for converting the bottom text area into a minimum bounding rectangle; A search module is used to search for a gray value jump point every 10 pixels along the downward direction of the minimum circumscribed rectangle to obtain a plurality of gray value jump points; A calculation module, used for calculating the distances from the lower left corner and the lower right corner of the minimum circumscribed rectangle to the reference line of the jewelry label, and taking the minimum value of the distances as the distance from the bottom text area to the reference line of the jewelry label; The first judgment module is used to judge whether the distance between the bottom text area and the reference line of the jewelry label is within 2-20 pixels. If so, it means that the position of the bottom text area in the area of ​​interest is qualified. If not, it means that the position of the bottom text area in the area of ​​interest is unqualified.

9. The jewelry information recognition and proofreading system based on deep learning according to claim 6, characterized in that: The data processing module comprises: A first filtering submodule is used to filter the text information in the data area of ​​the area of ​​interest in the pre-processed jewelry appraisal certificate image; The second filtering submodule is used to filter the text information of the data area of ​​the area of ​​interest in the pre-processed jewelry label image.

10. The jewelry information recognition and proofreading system based on deep learning according to claim 6, characterized in that: Also includes: An acquisition module, used to acquire the jewelry quality in the jewelry information of the jewelry appraisal certificate, the jewelry quality in the jewelry information of the jewelry label, and the actual quality of the jewelry; The second judgment module is used to judge whether the jewelry quality in the jewelry information of the jewelry appraisal certificate, the jewelry quality in the jewelry information of the jewelry label and the actual quality of the jewelry meet the following conditions: |M1-M3|≤d and |M2-M3|≤d; Among them, M1 represents the jewelry quality in the jewelry information of the jewelry appraisal certificate; M2 represents the jewelry quality in the jewelry information of the jewelry label; M3 represents the actual quality of the jewelry; d represents the allowable quality tolerance; if it is met, it proves that the quality of the jewelry appraisal certificate is qualified; if it is not met, it proves that the quality of the jewelry appraisal certificate is unqualified.

Images