Intelligent talent recommendation system based on talent database

Abstract

The invention belongs to the field of intelligent talent recommendation, and discloses an intelligent talent recommendation system based on a talent database, which comprises an input module, a database module, an input module and an intelligent recommendation module, the input module comprises a scanning unit; the scanning unit is used for shooting a paper resume to obtain a resume image, and is used for obtaining an electronic resume based on the resume image; the database module is used for storing the electronic resumes; the input module is used for a user to input talent demand texts; and the intelligent recommendation module is used for matching the electronic resumes of the recommended talents from the database module based on the talent demand text. According to the method, the probability of text information loss can be effectively reduced, so that a scanning and warehousing result is more accurate, and the effectiveness of recommended talents is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent talent recommendation, and in particular to an intelligent talent recommendation system based on a talent database. Background Art

[0002] A talent database refers to a database that stores a large number of resumes of people. When talent screening is required, the corresponding recommendation results can be automatically returned from the talent database by entering the screening conditions in the database management system. There are two sources of resumes in the talent database: one is the electronic resume transmitted by job seekers through emails, and the other is the paper resume received in offline presentations and other activities.

[0003] In the process of converting paper resumes into electronic resumes, the existing talent database system usually takes a photo of the resume, directly binarizes the entire image, and then uses a text recognition algorithm to recognize the binarized image to obtain the text content in the image. However, this processing method does not take into account the situation where the text is highlighted. Highlighting is generally a text format set by job seekers to highlight the importance of the text. Highlighting refers to generating a rectangular frame around the text. The pixels belonging to the text in the rectangular frame retain their original color, while the other pixels in the rectangular frame are changed to the highlight color selected by the user;

[0004] After highlighting, if the grayscale value of the pixels in the rectangular box that do not belong to the text is close to the grayscale value of the text, it is easy for the pixels in the rectangular box that do not belong to the text to be mistakenly retained due to the inaccurate binarization threshold. Moreover, because it is binarized, after retention, the grayscale value of the pixels in the rectangular box that do not belong to the text is the same as the grayscale value of the text, resulting in the loss of text information and the inability to perform effective text recognition. As a result, after image-based recognition, the resume obtained is incomplete, affecting the effectiveness of subsequent talent recommendations, that is, job seekers who should have been recommended may not be correctly recommended due to missing resume information. Summary of the invention

[0005] The purpose of the present invention is to disclose a talent intelligent recommendation system based on a talent database to solve the technical problems raised in the background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] The present invention provides a talent intelligent recommendation system based on a talent database, comprising an entry module, a database module, an input module and an intelligent recommendation module;

[0008] The input module includes a scanning unit;

[0009] The scanning unit is used to take a picture of the paper resume to obtain a resume image, and to obtain an electronic resume based on the resume image;

[0010] The database module is used to store electronic resumes;

[0011] The input module is used for users to input talent demand text;

[0012] The intelligent recommendation module is used to match the electronic resumes of recommended talents from the database module based on the talent demand text;

[0013] The electronic resume is obtained based on the resume image, including:

[0014] S1, grayscale the resume image to obtain a grayscale image;

[0015] S2, dividing the grayscale image into multiple regions with areas smaller than the grayscale image;

[0016] S3, performing binarization processing in each region respectively to obtain multiple binarized regions;

[0017] S4, stitching all the binary areas to obtain a binary image;

[0018] S5, performing connected domain detection on the binary image to obtain multiple connected domains;

[0019] S6, perform the following processing on each connected domain to obtain a binary image:

[0020] Determine whether the area of ​​the connected domain is greater than the adaptive area threshold. If so, optimize the connected domain. If not, keep the connected domain unchanged.

[0021] S7, recognizes the binary image and generates an electronic resume.

[0022] Preferably, the input module further includes a manual input unit;

[0023] The manual input unit is used for users to upload electronic resumes received through the network to the database module.

[0024] Preferably, the talent requirement text includes education background, age, major, years of work experience and project experience.

[0025] Preferably, matching electronic resumes of recommended talents from a database module based on the talent demand text includes:

[0026] Calculate the similarity between the talent demand text and each electronic resume respectively;

[0027] The top N electronic resumes with the highest similarity are used as electronic resumes of recommended talents.

[0028] Preferably, the similarity between the talent demand text and each electronic resume is calculated, including:

[0029] Calculate the Jaccard similarity or cosine similarity between the talent demand text and each electronic resume.

[0030] Preferably, grayscale processing is performed on the resume image to obtain a grayscale image, including:

[0031] The weighted average method is used to grayscale the resume image to obtain a grayscale image.

[0032] Preferably, the grayscale image is divided into a plurality of regions having an area smaller than the grayscale image, including:

[0033] The grayscale image is divided into multiple regions with smaller areas than the grayscale image by using multiple partitioning methods:

[0034] First calculation:

[0035] Divide the grayscale image into s regions of equal area;

[0036] Calculate the grayscale variation coefficient of each area separately;

[0037] The areas where the grayscale variation coefficient is less than the grayscale variation coefficient threshold are stored in the set U1; the areas where the grayscale variation coefficient is greater than or equal to the grayscale variation coefficient threshold are stored in the partition result set;

[0038] hth calculation:

[0039] Divide each region in U1 into s regions of equal area;

[0040] Calculate the grayscale variation coefficient of each area separately;

[0041] The area where the grayscale variation coefficient is less than the grayscale variation coefficient threshold is stored in the set U h ; Store the areas whose grayscale variation coefficient is greater than or equal to the grayscale variation coefficient threshold into the partition result set;

[0042] If U h If there is at least one element in the partition result set, the next calculation is performed; otherwise, the elements in the partition result set are used as the final region.

[0043] Preferably, the calculation formula of the grayscale variation coefficient is as follows:

[0044]

[0045] gryscl represents the grayscale variation coefficient, g irepresents the gradient value of pixel i, snu represents the set of pixels in the region, aveg represents the average gradient value of the pixels in snu, gma represents the maximum gradient value of the pixels in snu, ngry represents the total number of pixels in the grayscale image, ncetdom represents the total number of pixels in the connected domain containing the most pixels in the region; δ1 represents the gradient value change weight, and δ2 represents the connected domain weight.

[0046] Preferably, binarization is performed in each region to obtain multiple binarized regions, including:

[0047] The following calculations are performed for each region:

[0048] For region q, use the Otsu algorithm to calculate q and obtain the segmentation threshold qthre;

[0049] Get the grayscale histogram of q;

[0050] Get the set lwu of grayscale values ​​at the trough in the grayscale histogram;

[0051] Calculate the binarization threshold:

[0052]

[0053] binsthr represents the binarization threshold, sgry represents the preset grayscale value, and lthre represents the minimum grayscale value in lwu;

[0054] The grayscale values ​​of the pixels in the region whose grayscale values ​​are greater than the binary threshold are set to 255; the grayscale values ​​of the remaining pixels are set to 0, and the binary region corresponding to q is obtained.

[0055] Preferably, the calculation process of the adaptive area threshold includes:

[0056] Get the minimum enclosing rectangle of each connected domain respectively;

[0057] Twice the median of the areas of all minimum bounding rectangles is used as the adaptive threshold.

[0058] Beneficial effects:

[0059] Different from the existing talent recommendation system, the present invention does not directly perform binarization on the entire image during the process of scanning and storing paper resumes, but divides the image into multiple regions first, and then performs binarization in each region respectively, and then splices the binarized images obtained by the binarization process to obtain a binarized image, and then performs connected domain detection on the binarized image to identify the region where the pixel points belonging to the highlighted area are suspected to be mistakenly regarded as the pixel points belonging to the text, and finally optimizes the identified area. The present invention can effectively reduce the probability of text information loss, make the scanning and storage results more accurate, and thus improve the effectiveness of recommended talents. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0061] Figure 1 The present invention is a schematic diagram of a talent intelligent recommendation system based on a talent database.

[0062] Figure 2 A schematic diagram of the text highlighting effect. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0064] like Figure 1 In one embodiment shown, the present invention provides a talent intelligent recommendation system based on a talent database, including an entry module, a database module, an input module and an intelligent recommendation module;

[0065] The input module includes a scanning unit;

[0066] The scanning unit is used to take a picture of the paper resume to obtain a resume image, and to obtain an electronic resume based on the resume image;

[0067] The database module is used to store electronic resumes;

[0068] The input module is used for users to input talent demand text;

[0069] The intelligent recommendation module is used to match the electronic resumes of recommended talents from the database module based on the talent demand text;

[0070] The electronic resume is obtained based on the resume image, including:

[0071] S1, grayscale the resume image to obtain a grayscale image;

[0072] S2, dividing the grayscale image into multiple regions with areas smaller than the grayscale image;

[0073] S3, performing binarization processing in each region respectively to obtain multiple binarized regions;

[0074] S4, stitching all the binary areas to obtain a binary image;

[0075] S5, performing connected domain detection on the binary image to obtain multiple connected domains;

[0076] S6, perform the following processing on each connected domain to obtain a binary image:

[0077] Determine whether the area of ​​the connected domain is greater than the adaptive area threshold. If so, optimize the connected domain. If not, keep the connected domain unchanged.

[0078] S7, recognizes the binary image and generates an electronic resume.

[0079] In the above process, the image is first divided into multiple regions, and then binarization processing is performed in each region. After that, the binarized images obtained by the binarization processing are spliced ​​to obtain a binarized image, and then the binarized image is subjected to a connected domain detection to identify the region where the pixel points belonging to the highlighted area are suspected to be mistakenly regarded as the pixel points belonging to the text, and finally the identified region is optimized. The present invention can effectively reduce the probability of text information loss, make the results of scanning and warehousing more accurate, and thus improve the effectiveness of recommended talents.

[0080] Figure 2 Schematic diagram with highlighted text, in Figure 2In the figure, the blue area pointed by the arrow is the highlighted area. The highlighted area includes the pixels that belong to the text (black pixels) and the pixels that do not belong to the text (blue pixels).

[0081] In a resume, when a job seeker needs to highlight certain data, this form can be used to express it.

[0082] Preferably, the input module further includes a manual input unit;

[0083] The manual input unit is used for users to upload electronic resumes received through the network to the database module.

[0084] For example, electronic resumes received through WeChat, email, QQ, etc. These electronic resumes are generally in PDF format.

[0085] Preferably, the talent requirement text includes education background, age, major, years of work experience and project experience.

[0086] Educational background includes undergraduate, postgraduate, etc. Age can be an age range. Working experience refers to the cumulative number of years of work. Project experience refers to the experience of a specific type of project, such as server operation and maintenance projects.

[0087] Preferably, matching electronic resumes of recommended talents from a database module based on the talent demand text includes:

[0088] Calculate the similarity between the talent demand text and each electronic resume respectively;

[0089] The top N electronic resumes with the highest similarity are used as electronic resumes of recommended talents.

[0090] Specifically, N here can be set as needed, and the default value of N can be 10.

[0091] Preferably, the similarity between the talent demand text and each electronic resume is calculated, including:

[0092] Calculate the Jaccard similarity or cosine similarity between the talent demand text and each electronic resume.

[0093] Specifically, before calculating the similarity, if it is determined that any numerical requirement cannot be met, the similarity of the electronic resume is set to 0;

[0094] For example, when it is determined that the age item does not meet the age requirement in the talent demand text, the similarity of the electronic resume will be set to 0.

[0095] Preferably, grayscale processing is performed on the resume image to obtain a grayscale image, including:

[0096] The weighted average method is used to grayscale the resume image to obtain a grayscale image.

[0097] In addition, the resume image can be grayed out using methods such as the maximum value method.

[0098] Preferably, the grayscale image is divided into a plurality of regions having an area smaller than the grayscale image, including:

[0099] The grayscale image is divided into multiple regions with smaller areas than the grayscale image by using multiple partitioning methods:

[0100] First calculation:

[0101] Divide the grayscale image into s regions of equal area;

[0102] Calculate the grayscale variation coefficient of each area separately;

[0103] The areas where the grayscale variation coefficient is less than the grayscale variation coefficient threshold are stored in the set U1; the areas where the grayscale variation coefficient is greater than or equal to the grayscale variation coefficient threshold are stored in the partition result set;

[0104] hth calculation:

[0105] Divide each region in U1 into s regions of equal area;

[0106] Calculate the grayscale variation coefficient of each area separately;

[0107] The area where the grayscale variation coefficient is less than the grayscale variation coefficient threshold is stored in the set U h ; Store the areas whose grayscale variation coefficient is greater than or equal to the grayscale variation coefficient threshold into the partition result set;

[0108] If U h If there is at least one element in the partition result set, the next calculation is performed; otherwise, the elements in the partition result set are used as the final region.

[0109] Specifically, the present invention does not directly divide the grayscale image into multiple regions of the same area, because the images captured under different circumstances contain different contents. If the area of ​​the set region is too small, the calculation efficiency will be too low, and it is also easy to cause the subsequent binarization process to forcibly divide some pixels that originally do not belong to the text as pixels belonging to the text. Obviously, this is an erroneous binarization process.

[0110] Therefore, the present invention adopts a multiple calculation method to gradually judge whether each area needs to be partitioned for the next time, so that the partitioned areas contain pixels belonging to text and pixels that do not belong to text. In this way, the probability of an erroneous binarization process during binarization can be reduced.

[0111] Preferably, the calculation formula of the grayscale variation coefficient is as follows:

[0112]

[0113] gryscl represents the grayscale variation coefficient, g i represents the gradient value of pixel i, snu represents the set of pixels in the region, aveg represents the average gradient value of the pixels in snu, gma represents the maximum gradient value of the pixels in snu, ngry represents the total number of pixels in the grayscale image, ncetdom represents the total number of pixels in the connected domain containing the most pixels in the region; δ1 represents the gradient value change weight, and δ2 represents the connected domain weight.

[0114] The grayscale variation coefficient of the present invention is calculated from the two directions of gradient value and the largest connected domain respectively. Compared with only considering the gradient value as the grayscale variation coefficient, the present invention can avoid the area of ​​the region in the partition result set being too large, because this will easily lead to the subsequent calculation of the binarization threshold. The binarization threshold is too greatly affected by the grayscale value of the area that does not belong to the text, making the calculated threshold too small, and thus the pixel points of the highlighted area cannot be divided into the area that does not belong to the text, resulting in the pixel points in the highlighted area that do not belong to the text being mistakenly set to the same grayscale value as the pixel points belonging to the text, causing the text in the highlighted area to be lost after binarization.

[0115] Preferably, the gradient value change weight is The weight of the connected domain is

[0116] Preferably, the grayscale variation coefficient threshold is 0.3.

[0117] Since the grayscale variation coefficient threshold does not directly control the binarization process but only controls the partitioning process, it is possible to avoid the occurrence of an event where a fixed threshold is set, resulting in an inaccurate binarization result.

[0118] Preferably, binarization is performed in each region to obtain multiple binarized regions, including:

[0119] The following calculations are performed for each region:

[0120] For region q, use the Otsu algorithm to calculate q and obtain the segmentation threshold qthre;

[0121] Get the grayscale histogram of q;

[0122] Get the set lwu of grayscale values ​​at the trough in the grayscale histogram;

[0123] Calculate the binarization threshold:

[0124]

[0125] binsthr represents the binarization threshold, sgry represents the preset grayscale value, and lthre represents the minimum grayscale value in lwu;

[0126] The grayscale values ​​of the pixels in the region whose grayscale values ​​are greater than the binary threshold are set to 255; the grayscale values ​​of the remaining pixels are set to 0, and the binary region corresponding to q is obtained.

[0127] When calculating the binarization threshold, the present invention does not directly use the result calculated by the Otsu algorithm as the binarization threshold, because when the highlighted area exists, the threshold calculated by the Otsu algorithm may be too small, resulting in the pixel points in the highlighted area that do not belong to the text being erroneously assigned the same grayscale value as the pixel points belonging to the text after binarization. Therefore, the present invention obtains the minimum value of the grayscale value in the trough in the grayscale histogram, and then involves the minimum value in the calculation process of the binarization threshold, so that the binarization threshold can change adaptively with the change of the difference between qthre and lthre, effectively reducing the probability of the occurrence of the event that the binarization threshold is too small, and effectively improving the accuracy of the binarization result.

[0128] Specifically, the grayscale value at the trough is identified as follows:

[0129] For the gray value b, the height corresponding to the gray value b in the gray histogram is represented as hei b ;

[0130] In the gray value interval [bB, b+B], if there is no gray value with a height greater than hei in the gray histogram, b The gray value of , then b is the gray value at the trough.

[0131] Specifically, in the grayscale histogram, the height of b can be represented by the number of pixels in the region whose grayscale values ​​are equal to b. The greater the number, the higher the height.

[0132] Preferably, the value of B is 10.

[0133] Preferably, the calculation process of the adaptive area threshold includes:

[0134] Get the minimum enclosing rectangle of each connected domain respectively;

[0135] Twice the median of the areas of all minimum bounding rectangles is used as the adaptive threshold.

[0136] Specifically, after obtaining the connected domain, many local areas of text will also be regarded as a connected domain. Therefore, by obtaining the minimum circumscribed rectangle, the size of the area roughly corresponding to a single text can be obtained. When the text is highlighted, the area of ​​the highlighted area will be larger than the area of ​​the text. Therefore, by obtaining twice the median of the area, the area that may belong to the highlighted area can be identified from the binary image.

[0137] Preferably, the connected domain is optimized, including:

[0138] Get the set LK of coordinates of the pixel points in the connected domain;

[0139] In the grayscale image, obtain the area ALK composed of the pixel points corresponding to the coordinates in LK;

[0140] Calculate the standard deviation A1 of the grayscale values ​​of the pixels in ALK;

[0141] Calculate the average grayscale value A2 of the pixel points in ALK;

[0142] Determine whether |A1-A2| is less than the set gray value threshold. If so, keep the connected domain unchanged. If not, use the Otsu algorithm to calculate ALK and obtain the segmentation threshold ALKthr.

[0143] The grayscale values ​​of the pixels corresponding to the pixels in the connected domain whose grayscale values ​​in ALK are less than or equal to ALKthr are set to 0, and the grayscale values ​​of the remaining pixels in the connected domain are set to 255.

[0144] Specifically, during the optimization process, the present invention determines whether the connected domain belongs to the highlighted area containing text by calculating the absolute value of the difference between the standard deviation and the average value of the corresponding pixel points in the grayscale image of the pixel points in the connected domain. If not, the difference between the standard deviation and the average value should be the same under ideal conditions. However, due to the influence of uneven lighting during shooting, the present invention sets a gray value threshold for judgment. The smaller |A1-A2| is, the less likely it is that text is contained. Therefore, after determining that there is text, the present invention uses the Otsu algorithm to perform local segmentation on ALK, which can effectively separate the area that does not belong to the text from the area that belongs to the text in the highlighted area while avoiding the influence of the background, so that the gray value of the pixel points that originally belong to the text is retained in the binary image, and the pixel points that do not belong to the text are set to the same gray value as the background.

[0145] The grayscale value threshold can be set to 5.

[0146] Preferably, the binary image is recognized to generate an electronic resume, including:

[0147] Use the Tesseract OCR algorithm to perform text recognition in the binary image and record the coordinates of each recognized word;

[0148] A straight line detection algorithm is used to detect straight lines in the binary image, and straight lines whose length is less than the adaptive length threshold are removed to generate a blank resume form;

[0149] Fill the text recognized by the Tesseract OCR algorithm into the blank resume form according to its coordinates to generate an electronic resume.

[0150] For example, if the coordinates of the text in the binary image are (x, y), then its position in the blank resume form is also (x, y).

[0151] The established coordinate system can use the center of any character in the binary image as the coordinate origin.

[0152] Preferably, the process of obtaining the adaptive length threshold includes:

[0153] A straight line detection algorithm is used to detect straight lines from a binary image to obtain straight line segments in the binary image;

[0154] Use a clustering algorithm to cluster the lengths of all binary line segments. Set two cluster centers before clustering. The final output cluster centers are represented by L1 and L2 respectively. L1 is greater than L2.

[0155] The minimum length value in the cluster where L1 is located is used as the adaptive length threshold.

[0156] Since there are also straight line segments in text, after clustering, the length of the straight line segments belonging to the frame lines of the table on the resume can be effectively distinguished from the length of the straight line segments in the text.

[0157] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A talent intelligent recommendation system based on a talent database, characterized in that: It includes an entry module, a database module, an input module and an intelligent recommendation module; The input module includes a scanning unit; The scanning unit is used to take a picture of the paper resume to obtain a resume image, and to obtain an electronic resume based on the resume image; The database module is used to store electronic resumes; The input module is used for users to input talent demand text; The intelligent recommendation module is used to match the electronic resumes of recommended talents from the database module based on the talent demand text; The electronic resume is obtained based on the resume image, including: S1, grayscale the resume image to obtain a grayscale image; S2, dividing the grayscale image into multiple regions with areas smaller than the grayscale image; S3, performing binarization processing in each region respectively to obtain multiple binarized regions; S4, stitching all the binary areas to obtain a binary image; S5, performing connected domain detection on the binary image to obtain multiple connected domains; S6, perform the following processing on each connected domain to obtain a binary image: Determine whether the area of ​​the connected domain is greater than the adaptive area threshold. If so, optimize the connected domain. If not, keep the connected domain unchanged. S7, recognizes the binary image and generates an electronic resume.

2. According to claim 1, a talent intelligent recommendation system based on a talent database is characterized in that: The input module also includes a manual input unit; The manual input unit is used for users to upload electronic resumes received through the network to the database module.

3. According to claim 1, a talent intelligent recommendation system based on a talent database is characterized in that: The talent requirement text includes education background, age, major, years of work experience and project experience.

4. According to claim 1, a talent intelligent recommendation system based on a talent database is characterized in that: Match the electronic resumes of recommended talents from the database module based on the talent demand text, including: Calculate the similarity between the talent demand text and each electronic resume respectively; The top N electronic resumes with the highest similarity are used as electronic resumes of recommended talents.

5. According to claim 4, a talent intelligent recommendation system based on a talent database is characterized in that: Calculate the similarity between the talent demand text and each electronic resume, including: Calculate the Jaccard similarity or cosine similarity between the talent demand text and each electronic resume.

6. The talent intelligent recommendation system based on talent database according to claim 1, characterized in that: Grayscale the resume image to obtain a grayscale image, including: The weighted average method is used to grayscale the resume image to obtain a grayscale image.

7. The talent intelligent recommendation system based on talent database according to claim 1, characterized in that: Divide the grayscale image into multiple regions with smaller areas than the grayscale image, including: The grayscale image is divided into multiple regions with smaller areas than the grayscale image by using multiple partitioning methods: First calculation: Divide the grayscale image into s regions of equal area; Calculate the grayscale variation coefficient of each area respectively; The areas where the grayscale variation coefficient is less than the grayscale variation coefficient threshold are stored in the set U1; the areas where the grayscale variation coefficient is greater than or equal to the grayscale variation coefficient threshold are stored in the partition result set; The hth calculation: Divide each region in U1 into s regions of equal area; Calculate the grayscale variation coefficient of each area separately; The area where the grayscale variation coefficient is less than the grayscale variation coefficient threshold is stored in the set U h ; Store the areas whose grayscale variation coefficient is greater than or equal to the grayscale variation coefficient threshold into the partition result set; If U h If there is at least one element in the partition result set, the next calculation is performed; otherwise, the elements in the partition result set are used as the final region.

8. The talent intelligent recommendation system based on talent database according to claim 7, characterized in that: The calculation formula of grayscale variation coefficient is as follows: gryscl represents the grayscale variation coefficient, g i represents the gradient value of pixel i, snu represents the set of pixels in the region, aveg represents the average gradient value of the pixels in snu, gma represents the maximum gradient value of the pixels in snu, ngry represents the total number of pixels in the grayscale image, ncetdom represents the total number of pixels in the connected domain containing the most pixels in the region; δ1 represents the gradient value change weight, and δ2 represents the connected domain weight.

9. The talent intelligent recommendation system based on talent database according to claim 1, characterized in that: Binarization is performed in each region to obtain multiple binary regions, including: The following calculations are performed for each region: For region q, use the Otsu algorithm to calculate q and obtain the segmentation threshold qthre; Get the grayscale histogram of q; Get the set lwu of grayscale values ​​at the trough in the grayscale histogram; Calculate the binarization threshold: binsthr represents the binarization threshold, sgry represents the preset grayscale value, and lthre represents the minimum grayscale value in lwu; The grayscale values ​​of the pixels in the region whose grayscale values ​​are greater than the binary threshold are set to 255; the grayscale values ​​of the remaining pixels are set to 0, and the binary region corresponding to q is obtained.

10. The talent intelligent recommendation system based on talent database according to claim 1, characterized in that: The calculation process of the adaptive area threshold includes: Get the minimum enclosing rectangle of each connected domain respectively; Twice the median of the areas of all minimum bounding rectangles is used as the adaptive threshold.

Images