Precision marketing analysis system based on big data

Abstract

The invention belongs to the field of data processing, and discloses a precision marketing analysis system based on big data, which comprises an analysis terminal and an identity recognition terminal, the analysis terminal is used for acquiring a face image of an analyst and transmitting the face image to the identity recognition terminal; the identity recognition terminal comprises a noise reduction module and a recognition module; the noise reduction module is used for carrying out noise reduction on the face image to obtain a to-be-recognized image; the recognition module is used for recognizing the to-be-recognized image and judging whether the face recognition of the analyst is successful or not. According to the method, the noise reduction efficiency is higher, and compared with the first noise reduction mode adopted for all the pixel points, the edge information can be more effectively reserved. Therefore, the overall noise reduction time can be effectively shortened and the noise reduction efficiency is improved while the noise reduction effect is ensured, so that the customer type determination efficiency is improved and the marketing efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a precision marketing analysis system based on big data. Background Art

[0002] Marketing analysis based on big data refers to analyzing the customer's attribute data (such as gender, age, etc.) and behavior data (such as consumption records, etc.) to determine the customer's type, and then accurately marketing the customer based on the marketing strategy corresponding to the customer's type to improve the marketing effect. Since attribute data and behavior data are usually confidential data, data security needs to be guaranteed during the analysis process. In the prior art, when performing analysis, face recognition is usually performed on the identity of the analyst to prevent unauthorized personnel from obtaining the customer's attribute data and behavior data. In the process of face recognition, the same noise reduction algorithm is usually used directly for all pixels to reduce noise. If an algorithm that takes a long time but has a good noise reduction effect is used, the noise reduction efficiency will be low, resulting in a relatively long time to obtain the face recognition result, affecting the efficiency of determining the customer's type, thereby affecting the marketing efficiency; if an algorithm that takes a short time but has a poor noise reduction effect is used, although the noise reduction efficiency can be improved, due to the poor noise reduction effect, there are too many early points in the image, which may make the face recognition failure probability higher. Summary of the invention

[0003] The purpose of the present invention is to disclose a precision marketing analysis system based on big data to solve the technical problems pointed out in the background technology.

[0004] In order to achieve the above object, the present invention adopts the following technical solution:

[0005] The present invention provides a precision marketing analysis system based on big data, including an analysis terminal and an identity recognition terminal;

[0006] The analysis terminal is used to obtain the facial image of the analyst and transmit the facial image to the identity recognition terminal;

[0007] The identity recognition terminal includes a noise reduction module and an identification module;

[0008] The noise reduction module is used to reduce the noise of the face image in the following way to obtain the image to be recognized:

[0009] The first step is to obtain a set C1 of pixels that meet the skin color judgment rule and a set C2 of pixels that do not meet the skin color judgment rule in the face image;

[0010] In the second step, each pixel in C1 is judged as follows:

[0011] Let p1 represent the pixel in C1. If there is a pixel in C2 in the 8-neighborhood of p1, then p1 is taken as the pixel in set C3.

[0012] The third step is to obtain the noise reduction length corresponding to each pixel of C3;

[0013] The fourth step is to perform the following processing based on each pixel in C3:

[0014] Let p3 represent the pixel point in C3, and let xp3 and yp3 represent the horizontal coordinate and vertical coordinate of p3 respectively;

[0015] The pixel points with ordinate yp3 and abscissa in the interval [xp3-zp3, xp3+zp3] are stored in set C4, where zp3 represents the noise reduction length corresponding to p3;

[0016] Step 5: Use the first noise reduction method to reduce the noise of the pixels in C1 and C2 to obtain a set E1; use the second noise reduction method to reduce the noise of the pixels in C4 to obtain a set E2;

[0017] Step 6: Process the pixels in E1 as follows:

[0018] Let pe2 represent the pixel in E2, pe1 represent the pixel in E1 with the same coordinates as pe2, and Gpe1 represent the gray value of pe1;

[0019] Use the following formula to optimize the gray value of pe1:

[0020]

[0021] cpe1 represents the gray value of pe1 after optimization, and Gpe2 represents the gray value of pe2;

[0022] Step 7: The pixel points in E1 form the image to be recognized;

[0023] The recognition module is used to identify the image to be identified and determine whether the analyst's face recognition is successful.

[0024] Preferably, it also includes a database terminal;

[0025] The recognition module is used to send the identity verification code to the analysis terminal and the database terminal respectively after the face recognition of the analyst is successful;

[0026] The analysis terminal is used to send the authentication code and the download request input by the analyst to the database terminal;

[0027] The database terminal is used to determine whether the verification code sent by the analysis terminal is consistent with the verification code sent by the identification module. If so, it responds to the download request and transmits data to the analysis terminal.

[0028] Preferably, the analysis terminal is also used to input the data transmitted from the database terminal into the classification model for classification to obtain the type of customer.

[0029] Preferably, it also includes a marketing terminal;

[0030] The marketing terminal is used to send marketing advertisements to customers based on their types.

[0031] Preferably, obtaining a set C1 of pixels that meet the skin color determination rule and a set C2 of pixels that do not meet the skin color determination rule in the face image includes:

[0032] Use pf to represent the pixel points in the face image;

[0033] Determine whether pf meets the following skin color judgment rules. If so, store pf in set C1; if not, store pf in C2:

[0034]

[0035] H pf , S pf and V pf They respectively represent the values ​​of the H component, S component, and V component of pf in the HSV color space.

[0036] Preferably, the calculation formula for the noise reduction length corresponding to the pixel point in C3 is:

[0037]

[0038] zp3 represents the denoising length corresponding to the pixel point p3 in C3, nep3 represents the set of pixels in a square window with a length of 5 and centered on p3, Gi represents the grayscale value of the pixel point i in nep3, avG represents the average grayscale value of the pixel points in nep3, nC3 represents the number of pixels in nep3 belonging to the set C3, λ represents the calculation weight; zl is the preset length.

[0039] Preferably, the first noise reduction method is used to reduce noise on the pixels in C1 and C2, and the set E1 is obtained including:

[0040] Use Gaussian filtering algorithm to reduce the noise of pixels in C1 to obtain set D1;

[0041] Use the guided filtering algorithm to reduce the noise of the pixels in C2 to obtain the set D2;

[0042] The set E1 is composed of the sets D1 and D2.

[0043] Preferably, the second noise reduction method is used to reduce the noise of the pixels in C4 to obtain a set E2, including:

[0044] For pixel pc4 in C4, use the following formula to reduce noise:

[0045]

[0046]

[0047] aGpc4 represents the denoising middle value of pc4, Gpc4 represents the gray value of pc4 before denoising, Gk represents the gray value of pixel k, nepc4 represents the set of pixels in a square window with a length of 5 centered on pc4, δ is a preset positive integer, represents the first noise reduction control value, represents the second noise reduction control value; fs k represents the standard deviation of the grayscale values ​​of pixels in a square window with a length of 5 and centered at k, fs pc4 represents the standard of the grayscale value of the pixels in the square window with a length of 5 and centered on pc4; Gm represents the maximum grayscale value of the pixels in nepc4; dpc4k is the number of pixels passed by the straight line segment between k and pc4;

[0048]

[0049] bGpc4 represents the grayscale value after pc4 noise reduction.

[0050] Preferably, identifying the image to be identified and judging whether the face recognition of the analyst is successful include:

[0051] Calculate the similarity between the image to be identified and each image in the database. If the maximum similarity is greater than the set similarity, it means that the analyst's face recognition is successful;

[0052] The images in the database are face images of all analysts who have download permissions.

[0053] Preferably, for an image s in the database, the similarity between it and the image to be identified is calculated as follows:

[0054] Use the face feature extraction algorithm to obtain the feature vector vec1 of s;

[0055] Use the face feature extraction algorithm to obtain the feature vector vec2 of the image to be recognized;

[0056] The cosine similarity between vec1 and vec2 is taken as the similarity between s and the image to be recognized.

[0057] Beneficial effects:

[0058] The present invention improves the existing face recognition process. In the denoising step of face recognition, denoising is no longer performed based on only one denoising algorithm. Instead, it is first performed based on compliance with the judgment rule and obtaining the sets C1 and C2, and then based on C1 and C2, a set C4 of pixels near the boundary of C1 and C2 is obtained. Then, two different denoising methods are used to denoise the pixels of C1 and C2 and the pixels in C4, respectively, to obtain sets E1 and E2. Finally, the grayscale value of the pixels of E1 is optimized based on the pixels of E2, so that the efficiency of denoising can be effectively improved while effectively retaining the edge information of the image, because the pixels in C4 only account for a very small part of the total number of C1 and C2. Therefore, compared with the second denoising method for all pixels, the denoising efficiency of the present invention is higher, and compared with the first denoising method for all pixels, the edge information can be more effectively retained. Therefore, the present invention can effectively shorten the overall denoising time while ensuring the effect of denoising, improve the efficiency of denoising, thereby improving the efficiency of determining the type of customers and improving the efficiency of marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] 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.

[0060] Figure 1 This is a first schematic diagram of a precision marketing analysis system based on big data according to the present invention.

[0061] Figure 2 This is a second schematic diagram of a precision marketing analysis system based on big data according to the present invention. DETAILED DESCRIPTION

[0062] 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.

[0063] like Figure 1 In one embodiment shown, the present invention provides a precision marketing analysis system based on big data, including an analysis terminal and an identity recognition terminal;

[0064] The analysis terminal is used to obtain the facial image of the analyst and transmit the facial image to the identity recognition terminal;

[0065] The identity recognition terminal includes a noise reduction module and an identification module;

[0066] The noise reduction module is used to reduce the noise of the face image in the following way to obtain the image to be recognized:

[0067] The first step is to obtain a set C1 of pixels that meet the skin color judgment rule and a set C2 of pixels that do not meet the skin color judgment rule in the face image;

[0068] In the second step, each pixel in C1 is judged as follows:

[0069] Let p1 represent the pixel in C1. If there is a pixel in C2 in the 8-neighborhood of p1, then p1 is taken as the pixel in set C3.

[0070] The third step is to obtain the noise reduction length corresponding to each pixel of C3;

[0071] The fourth step is to perform the following processing based on each pixel in C3:

[0072] Let p3 represent the pixel point in C3, and let xp3 and yp3 represent the horizontal coordinate and vertical coordinate of p3 respectively;

[0073] The pixel points with ordinate yp3 and abscissa in the interval [xp3-zp3, xp3+zp3] are stored in set C4, where zp3 represents the noise reduction length corresponding to p3;

[0074] Step 5: Use the first noise reduction method to reduce the noise of the pixels in C1 and C2 to obtain a set E1; use the second noise reduction method to reduce the noise of the pixels in C4 to obtain a set E2;

[0075] Step 6: Process the pixels in E1 as follows:

[0076] Let pe2 represent the pixel in E2, pe1 represent the pixel in E1 with the same coordinates as pe2, and Gpe1 represent the gray value of pe1;

[0077] Use the following formula to optimize the gray value of pe1:

[0078]

[0079] cpe1 represents the gray value of pe1 after optimization, and Gpe2 represents the gray value of pe2;

[0080] By optimizing pe1, the detail information in pe2 can be introduced, so that the edge of the image composed of pixels based on E1 is clearer;

[0081] Step 7: The pixel points in E1 form the image to be recognized;

[0082] The recognition module is used to identify the image to be identified and determine whether the analyst's face recognition is successful.

[0083] The above implementation method improves the existing face recognition process. In the denoising step of face recognition, denoising is no longer performed based on only one denoising algorithm. Instead, it is first performed based on the judgment rule and obtaining the sets C1 and C2, and then based on C1 and C2, a set C4 of pixels near the boundary of C1 and C2 is obtained. Then, two different denoising methods are used to denoise the pixels of C1 and C2 and the pixels in C4, respectively, to obtain sets E1 and E2. Finally, the grayscale value of the pixels of E1 is optimized based on the pixels of E2, so that the efficiency of denoising can be effectively improved while effectively retaining the edge information of the image, because the pixels in C4 only account for a very small part of the total number of C1 and C2. Therefore, compared with the second denoising method for all pixels, the denoising efficiency of the present invention is higher, and compared with the first denoising method for all pixels, the edge information can be more effectively retained. Therefore, the present invention can effectively shorten the overall denoising time while ensuring the effect of denoising, improve the efficiency of denoising, thereby improving the efficiency of determining the type of customers and improving the efficiency of marketing.

[0084] Preferably, if Figure 2 , also includes a database terminal;

[0085] The recognition module is used to send the identity verification code to the analysis terminal and the database terminal respectively after the face recognition of the analyst is successful;

[0086] The analysis terminal is used to send the authentication code and the download request input by the analyst to the database terminal;

[0087] The database terminal is used to determine whether the verification code sent by the analysis terminal is consistent with the verification code sent by the identification module. If so, it responds to the download request and transmits data to the analysis terminal.

[0088] Furthermore, the identity verification code may be a randomly generated positive integer of a preset length.

[0089] For example, a randomly generated positive integer with a length of 20 digits.

[0090] Furthermore, the download request includes the identity information of the client and the type of data to be downloaded. The type of data includes at least one of attribute data and behavior data.

[0091] Attribute data includes age, address, etc.; behavioral data includes website click records, consumption records, etc.

[0092] Identity information includes mobile phone number, ID card number and other information that can be matched with the customer.

[0093] Further, in response to the download request, transmitting data to the analysis terminal includes:

[0094] The corresponding data is obtained according to the identity information in the download request and the type of data to be downloaded, and the obtained data is transmitted to the analysis terminal.

[0095] Preferably, the analysis terminal is also used to input the data transmitted from the database terminal into the classification model for classification to obtain the type of customer.

[0096] Furthermore, the classification model may be a SVM model, a random forest model, or the like.

[0097] Furthermore, the types of customers can be set according to actual needs. For example, for a shopping platform, the types of customers include potential customers, new customers, returning customers, loyal customers, high-value customers, etc.

[0098] Potential customers: People who are interested in a product or service but have not yet purchased it. They may be browsing the website, looking at products but not placing an order.

[0099] New customers: users who make purchases on the platform for the first time. They may be curious about the platform and products and want to explore them.

[0100] Repeat customers: Customers who have purchased on the platform before may have a preference for certain brands or products and are prone to making a second purchase.

[0101] Loyal customers: Users who frequently purchase on the platform, have strong loyalty to the brand or platform, and usually participate in promotions or membership programs.

[0102] High-value customers: users with higher spending amounts, usually returning or loyal customers, who contribute more to the platform.

[0103] Preferably, it also includes a marketing terminal;

[0104] The marketing terminal is used to send marketing advertisements to customers based on their types.

[0105] For potential users, the marketing advertisement can be a website for redeeming coupons for the products they viewed.

[0106] For new customers, marketing ads can be a website for redeeming coupons for all product categories.

[0107] For repeat customers and loyal customers, marketing ads can be a website for redeeming consumer points or platform membership ads;

[0108] For high-value customers, marketing advertising can be limited edition or customized products that meet their special needs.

[0109] Preferably, obtaining a set C1 of pixels that meet the skin color determination rule and a set C2 of pixels that do not meet the skin color determination rule in the face image includes:

[0110] Use pf to represent the pixel points in the face image;

[0111] Determine whether pf meets the following skin color judgment rules. If so, store pf in set C1; if not, store pf in C2:

[0112]

[0113] H pf , S pf and V pf They respectively represent the values ​​of the H component, S component, and V component of pf in the HSV color space.

[0114] Preferably, the calculation formula for the noise reduction length corresponding to the pixel point in C3 is:

[0115]

[0116] zp3 represents the denoising length corresponding to the pixel point p3 in C3, nep3 represents the set of pixels in a square window with a length of 5 and centered on p3, Gi represents the grayscale value of the pixel point i in nep3, avG represents the average grayscale value of the pixel points in nep3, nC3 represents the number of pixels in nep3 belonging to the set C3, λ represents the calculation weight; zl is the preset length.

[0117] By obtaining the noise reduction length corresponding to each pixel point respectively, the number of pixels in C4 can be effectively controlled, because if a fixed noise reduction length is set, then when the grayscale values ​​of the pixels around p3 are very close, and the number of pixels belonging to C3 around p3 is small, the second noise reduction method will be used to reduce the noise of too many pixels around p3, and the area around p3 is originally an area with a small change in grayscale value, which will affect the overall noise reduction efficiency; when the grayscale values ​​of the pixels around p3 are very different, and the number of pixels belonging to C3 around p3 is large, the second noise reduction method will be used to reduce the noise of too few pixels around p3, which is not conducive to the preservation of the final edge information. Therefore, the present invention calculates zp3 based on the grayscale values ​​of the pixels around p3 and the number of pixels belonging to C3, so that the final noise reduction result can retain more edge information while maintaining the noise reduction efficiency.

[0118] Further, the calculation weight may be 0.53 and the preset length may be 5.

[0119] Preferably, the first noise reduction method is used to reduce noise on the pixels in C1 and C2, and the set E1 is obtained including:

[0120] Use Gaussian filtering algorithm to reduce the noise of pixels in C1 to obtain set D1;

[0121] Use the guided filtering algorithm to reduce the noise of the pixels in C2 to obtain the set D2;

[0122] The set E1 is composed of the sets D1 and D2.

[0123] Specifically, the time complexity of the Gaussian filtering algorithm and the guided filtering algorithm is roughly O(N), where N represents the total number of pixels.

[0124] Therefore, the noise reduction efficiency of the Gaussian filter algorithm and the guided filter algorithm is higher. The guided filter can retain more edge details compared with the high and low filters because it introduces additional information about the mean and variance of the image. Since the pixels in C1 are pixels that meet the skin color judgment rules, and the grayscale values ​​inside the skin area basically change smoothly, the Gaussian filter algorithm with fast calculation speed is used to reduce noise in the skin area;

[0125] C2 contains more pixels that may belong to the edge of the image. Therefore, the present invention uses a guided filtering algorithm that has a better effect of retaining edge information to reduce noise.

[0126] Preferably, the second noise reduction method is used to reduce the noise of the pixels in C4 to obtain a set E2, including:

[0127] For pixel pc4 in C4, use the following formula to reduce noise:

[0128]

[0129] aGpc4 represents the denoising middle value of pc4, Gpc4 represents the gray value of pc4 before denoising, Gk represents the gray value of pixel k, nepc4 represents the set of pixels in a square window with a length of 5 centered on pc4, δ is a preset positive integer, represents the first noise reduction control value, represents the second noise reduction control value; fs k represents the standard deviation of the grayscale values ​​of pixels in a square window with a length of 5 and centered at k, fs pc4 represents the standard of the grayscale value of the pixels in the square window with a length of 5 and centered on pc4; Gm represents the maximum grayscale value of the pixels in nepc4; dpc4k is the number of pixels passed by the straight line segment between k and pc4;

[0130]

[0131] bGpc4 represents the grayscale value after pc4 noise reduction.

[0132] The time complexity of the second noise reduction method of the present invention is greater than O(N×5 2 ), therefore, the calculation speed of the second denoising method is much slower than that of the first denoising method. However, the above-mentioned denoising method of the present invention not only takes into account the difference in grayscale values ​​between pixels and the number of pixels passed by the straight line segments between pixels, but also takes into account the standard deviation of the grayscale values ​​of pixels in the neighborhood of pc4 and k, which is beneficial for making pixels with higher comprehensive similarities in many aspects provide greater influence in the denoising process. In this way, the probability of the situation that the pixels themselves are similar but the surroundings are too different, which is easy to occur when only the similarity of the pixels themselves is considered for denoising, can be reduced. In this way, the denoising effect can be further improved.

[0133] Furthermore, the preset positive integer may be 3.

[0134] Furthermore, the value range of the first noise reduction control value is [10, 40]; the value range of the second noise reduction control value is [2, 9].

[0135] Preferably, the first noise reduction control value is 25, and the second noise reduction control value is 5.

[0136] Preferably, identifying the image to be identified and judging whether the face recognition of the analyst is successful include:

[0137] Calculate the similarity between the image to be identified and each image in the database. If the maximum similarity is greater than the set similarity, it means that the analyst's face recognition is successful;

[0138] The images in the database are face images of all analysts who have download permissions.

[0139] Specifically, the similarity is set to 0.7.

[0140] Preferably, for an image s in the database, the similarity between it and the image to be identified is calculated as follows:

[0141] Use the face feature extraction algorithm to obtain the feature vector vec1 of s;

[0142] Use the face feature extraction algorithm to obtain the feature vector vec2 of the image to be recognized;

[0143] The cosine similarity between vec1 and vec2 is taken as the similarity between s and the image to be recognized.

[0144] Specifically, the face feature extraction algorithm may be LBPH, Haar feature extraction algorithm, etc.

[0145] 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 precision marketing analysis system based on big data, characterized in that: Including analysis terminal and identity recognition terminal; The analysis terminal is used to obtain the facial image of the analyst and transmit the facial image to the identity recognition terminal; The identity recognition terminal includes a noise reduction module and an identification module; The noise reduction module is used to reduce the noise of the face image in the following way to obtain the image to be recognized: The first step is to obtain a set C1 of pixels that meet the skin color judgment rule and a set C2 of pixels that do not meet the skin color judgment rule in the face image; In the second step, each pixel in C1 is judged as follows: Let p1 represent the pixel in C1. If there is a pixel in C2 in the 8-neighborhood of p1, then p1 is taken as the pixel in set C3. The third step is to obtain the noise reduction length corresponding to each pixel of C3; The fourth step is to perform the following processing based on each pixel in C3: Let p3 represent the pixel point in C3, and let xp3 and yp3 represent the horizontal coordinate and vertical coordinate of p3 respectively; The pixel points with ordinate yp3 and abscissa in the interval [xp3-zp3, xp3+zp3] are stored in set C4, where zp3 represents the noise reduction length corresponding to p3; Step 5: Use the first noise reduction method to reduce the noise of the pixels in C1 and C2 to obtain a set E1; use the second noise reduction method to reduce the noise of the pixels in C4 to obtain a set E2; Step 6: Process the pixels in E1 as follows: Let pe2 represent the pixel in E2, pe1 represent the pixel in E1 with the same coordinates as pe2, and Gpe1 represent the gray value of pe1; Use the following formula to optimize the gray value of pe1: cpe1 represents the gray value of pe1 after optimization, and Gpe2 represents the gray value of pe2; Step 7: The pixel points in E1 form the image to be recognized; The recognition module is used to identify the image to be identified and determine whether the analyst's face recognition is successful.

2. The precision marketing analysis system based on big data according to claim 1, characterized in that: Also includes database terminal; The recognition module is used to send the identity verification code to the analysis terminal and the database terminal respectively after the face recognition of the analyst is successful; The analysis terminal is used to send the authentication code and the download request input by the analyst to the database terminal; The database terminal is used to determine whether the verification code sent by the analysis terminal is consistent with the verification code sent by the identification module. If so, it responds to the download request and transmits data to the analysis terminal.

3. The precision marketing analysis system based on big data according to claim 2 is characterized in that: The analysis terminal is also used to input the data transmitted from the database terminal into the classification model for classification to obtain the customer type.

4. The precision marketing analysis system based on big data according to claim 3 is characterized in that: It also includes marketing terminals; The marketing terminal is used to send marketing advertisements to customers based on their types.

5. The precision marketing analysis system based on big data according to claim 1, characterized in that: Obtaining a set C1 of pixels that meet the skin color judgment rule and a set C2 of pixels that do not meet the skin color judgment rule in the face image, including: Use pf to represent the pixel points in the face image; Determine whether pf meets the following skin color judgment rules. If so, store pf in set C1; if not, store pf in C2: H pf , S pf and V pf They respectively represent the values ​​of the H component, S component, and V component of pf in the HSV color space.

6. The precision marketing analysis system based on big data according to claim 1, characterized in that: The calculation formula for the noise reduction length corresponding to the pixel point in C3 is: zp3 represents the denoising length corresponding to the pixel point p3 in C3, nep3 represents the set of pixels in a square window with a length of 5 and centered on p3, Gi represents the grayscale value of the pixel point i in nep3, avG represents the average grayscale value of the pixel points in nep3, nC3 represents the number of pixels in nep3 belonging to the set C3, λ represents the calculation weight; zl is the preset length.

7. The precision marketing analysis system based on big data according to claim 1, characterized in that: The first noise reduction method is used to reduce the noise of the pixels in C1 and C2, and the obtained set E1 includes: Use Gaussian filtering algorithm to reduce the noise of pixels in C1 to obtain set D1; Use the guided filtering algorithm to reduce the noise of the pixels in C2 to obtain the set D2; The set E1 is composed of the sets D1 and D2.

8. The precision marketing analysis system based on big data according to claim 1, characterized in that: The second noise reduction method is used to reduce the noise of the pixels in C4 to obtain a set E2, including: For pixel pc4 in C4, use the following formula to reduce noise: aGpc4 represents the denoising middle value of pc4, Gpc4 represents the gray value of pc4 before denoising, Gk represents the gray value of pixel k, nepc4 represents the set of pixels in a square window with a length of 5 centered on pc4, δ is a preset positive integer, represents the first noise reduction control value, represents the second noise reduction control value; fs k represents the standard deviation of the grayscale values ​​of pixels in a square window with a length of 5 and centered at k, fs pc4 represents the standard of the grayscale value of the pixels in the square window with a length of 5 and centered on pc4; Gm represents the maximum grayscale value of the pixels in nepc4; dpc4k is the number of pixels passed by the straight line segment between k and pc4; bGpc4 represents the grayscale value after pc4 noise reduction.

9. The precision marketing analysis system based on big data according to claim 1, characterized in that: Identify the image to be identified and determine whether the analyst's face recognition is successful, including: Calculate the similarity between the image to be identified and each image in the database. If the maximum similarity is greater than the set similarity, it means that the analyst's face recognition is successful; The images in the database are face images of all analysts who have download permissions.

10. The precision marketing analysis system based on big data according to claim 1, characterized in that: For an image s in the database, the similarity between it and the image to be identified is calculated as follows: Use the face feature extraction algorithm to obtain the feature vector vec1 of s; Use the face feature extraction algorithm to obtain the feature vector vec2 of the image to be recognized; The cosine similarity between vec1 and vec2 is taken as the similarity between s and the image to be recognized.

Images