Personalized virtual fitting recommendation method and system based on user body type data

Abstract

The invention discloses a personalized virtual fitting recommendation method and system based on user body type data, and belongs to the technical field of computers. By acquiring multi-view images of a user in different postures, body characteristics are accurately captured, and a highly personalized user body model is constructed. And in combination with commodity evaluation and refunding and exchanging reasons in historical purchase records, intelligent clothing recommendation is realized, and the recommended clothing is ensured to conform to user preferences and body types. The advanced feature extraction, fusion and transformation technology is adopted in the human body-clothing fusion network, noise introduction and Prompt semantic description are combined, Mask sampling and Controlnet control are carried out, and a highly vivid virtual fitting effect is generated. According to the method, the personalization and the accuracy of virtual fitting are improved, more convenient, efficient and immersive shopping experience is provided for the user, and remarkable technical advantages and innovation values are shown.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more particularly to a personalized virtual fitting recommendation method and system based on user body shape data. Background Art

[0002] With the rapid development of e-commerce and digital technology, virtual fitting technology, as an innovative shopping method, is gradually gaining widespread attention from consumers and merchants.

[0003] However, the existing virtual fitting technology still has some obvious deficiencies in practical applications, which limit its further promotion and application. Existing virtual fitting technology mostly relies on static models and simple image processing, which is difficult to accurately reflect the real wearing effect. In addition, some virtual fitting systems lack personalized recommendation functions and cannot make accurate recommendations based on the user's personalized needs and body shape data, resulting in the recommended clothing not meeting the user's actual needs.

[0004] Therefore, how to propose a personalized virtual fitting recommendation method and system based on user body shape data to provide users with more accurate and personalized virtual fitting services is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a personalized virtual fitting recommendation method and system based on user body shape data, which not only improves the personalization and accuracy of virtual fitting, but also provides users with a more convenient, efficient and immersive shopping experience.

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

[0007] In one aspect, the present invention discloses a personalized virtual fitting recommendation method based on user body shape data, comprising the following steps:

[0008] Obtain multi-view human body images of the user and construct a human body model of the user;

[0009] According to user needs, analyze the user's historical purchase records to obtain recommended clothing models;

[0010] Determine the recommended clothing model size based on the user body model to obtain a clothing model to be tested;

[0011] A fusion operation is performed on the user body model and the clothing model to be tried on to generate a virtual fitting effect.

[0012] Preferably, obtaining multi-view human body images of a user and constructing a human body model of the user includes:

[0013] Acquire a first human body image of a user in a first posture;

[0014] identifying loose parts of human clothing and tight parts of the human body in the first human body image;

[0015] Instructing the user to make at least one predetermined second posture so that the loose part of the human body clothing becomes a compact part, and obtaining at least one second human body image in the second posture of the user;

[0016] Extracting human body feature parameters according to the first human body image and at least one second human body image;

[0017] A user body model is created according to the body feature parameters.

[0018] Preferably, according to user needs, the user's historical purchase records are analyzed to obtain a recommended clothing model, including:

[0019] Filter historical purchase records that meet user needs, wherein the historical purchase records include product evaluation information and return and exchange reason information;

[0020] From the historical purchase records that meet the user's needs, retain the product information whose product evaluation information exceeds the preset target, and remove some product information based on the return reason information to obtain reference product information;

[0021] The reference product information is analyzed to make clothing recommendations, and the corresponding clothing models are used as recommended clothing models.

[0022] Preferably, performing a fusion operation on the user body model and the clothing model to be tried on to generate a virtual fitting effect includes:

[0023] Constructing a human-clothing fusion network, the human-clothing fusion network includes a first feature extraction network, a second feature extraction network, a feature fusion network, and a transformation network; the first feature extraction network obtains human skeleton posture feature data based on a user human body model, the second feature extraction network obtains clothing feature data based on a clothing model to be tested, the feature fusion network performs feature fusion processing on the human skeleton posture feature data and the clothing feature data to obtain a preliminary fusion feature image and inputs the image into the transformation network to obtain a mixed feature image;

[0024] Introduce noise and generate the target dressing effect with vector representation based on the Prompt semantic description encoder;

[0025] Based on a sampling algorithm with Mask, the mixed feature image and the target dressing effect are iteratively sampled through the controlnet control network to obtain a preliminary dressing result;

[0026] The user body model, the clothing model to be tried on and the preliminary dressing result are encoded and mixed in sequence through the GAN network to generate a virtual fitting effect.

[0027] Preferably, the first feature extraction network and the second feature extraction network both include a first network layer, a second network layer and a third network layer; the first network layer, the second network layer and the third network layer are connected in sequence, and the first network layer, the second network layer and the third network layer all include a convolutional layer, a normalization layer and a loss function.

[0028] Preferably, the transformation network performs thin plate spline transformation processing on the preliminary fused feature image to obtain a mixed feature image.

[0029] On the other hand, the present invention also proposes a personalized virtual fitting recommendation system based on user body shape data, comprising:

[0030] A human body model building module is used to obtain multi-view human body images of the user and build a human body model of the user;

[0031] The clothing screening module is used to analyze the user's historical purchase records according to user needs and obtain recommended clothing models;

[0032] A clothing model acquisition module to be tested, used to determine the recommended clothing model size based on the user body model to obtain the clothing model to be tested;

[0033] The fitting effect display module is used to perform a fusion operation on the user body model and the clothing model to be tried on to generate a virtual fitting effect.

[0034] Preferably, the human body model building module includes:

[0035] A first image acquisition unit, used to acquire a first human body image of a user in a first posture;

[0036] a recognition unit, configured to recognize loose parts of human clothing and tight parts of the human body in the first human body image;

[0037] A second image acquisition unit is used to guide the user to make at least one predetermined second posture, so that the loose part of the human body clothing becomes a compact part, and acquire at least one second human body image in the second posture of the user;

[0038] A parameter extraction unit, configured to extract human body characteristic parameters according to the first human body image and at least one second human body image;

[0039] The user human body model generating unit is used to create a user human body model according to the human body feature parameters.

[0040] Preferably, the clothing screening module includes:

[0041] A preliminary screening unit, used to screen historical purchase records that meet user needs, wherein the historical purchase records include product evaluation information and return and exchange reason information;

[0042] A reference information acquisition unit is used to retain the product information whose product evaluation information exceeds the preset target from the historical purchase records that meet the user's needs, and to eliminate part of the product information based on the return reason information to obtain reference product information;

[0043] The clothing recommendation unit is used to analyze the reference commodity information to make clothing recommendations and use the corresponding clothing model as the recommended clothing model.

[0044] Preferably, the fitting effect display module includes:

[0045] A fusion network construction unit is used to construct a human-clothing fusion network, wherein the human-clothing fusion network includes a first feature extraction network, a second feature extraction network, a feature fusion network, and a transformation network; the first feature extraction network obtains human skeleton posture feature data based on a user human body model, the second feature extraction network obtains clothing feature data based on a clothing model to be tested, the feature fusion network performs feature fusion processing on the human skeleton posture feature data and the clothing feature data to obtain a preliminary fusion feature image and inputs the image into the transformation network to obtain a mixed feature image;

[0046] An effect generation unit, used for introducing noise and generating a target dressing effect with a vector representation based on a Prompt semantic description encoder;

[0047] A dressing unit, for iteratively sampling the mixed feature image and the target dressing effect through a controlnet control network based on a sampling algorithm with a mask, to obtain a preliminary dressing result;

[0048] The fitting effect display unit is used to encode and mix the user body model, the clothing model to be tried on and the preliminary dressing result in sequence through a GAN network to generate a virtual fitting effect.

[0049] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a personalized virtual fitting recommendation method and system based on user body shape data. By acquiring multi-view images of users in different postures, it can accurately capture body shape features and build a highly personalized user body model. Combined with the product evaluation and return reasons in the historical purchase records, intelligent clothing recommendations are realized to ensure that the recommended clothing is in line with user preferences and fits the body shape. The human-clothing fusion network adopts advanced feature extraction, fusion and transformation technology, combined with noise introduction and Prompt semantic description, as well as Mask sampling and controlnet control, to generate a highly realistic virtual fitting effect. The present invention not only improves the personalization and accuracy of virtual fitting, but also provides users with a more convenient, efficient and immersive shopping experience, showing significant technical advantages and innovative value. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0051] Figure 1 A method flow chart of a personalized virtual fitting recommendation method based on user body shape data provided by the present invention;

[0052] Figure 2 This is a system architecture diagram of the personalized virtual fitting recommendation system based on user body shape data provided by the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] On the one hand, the embodiment of the present invention discloses a personalized virtual fitting recommendation method based on user body shape data, referring to Figure 1 , the method comprises the following steps:

[0055] S1. Obtain multi-view human body images of the user and construct a human body model of the user, including:

[0056] S11. Obtain a first human body image of a user in a first posture.

[0057] S12. Identify loose parts of the human body's clothing and tight parts of the human body in the first human body image.

[0058] S13. Instruct the user to make at least one predetermined second posture so that the loose part of the human body clothing becomes a compact part, and obtain at least one second human body image in the second posture of the user.

[0059] S14. Extracting human body feature parameters according to the first human body image and at least one second human body image.

[0060] S15. Create a user body model according to the body feature parameters.

[0061] In this embodiment, the first human body image and the second human body image are both RGBD images. The human body RGBD image is a depth image including depth information. The depth image can be acquired by a depth camera. The depth camera can be a depth camera based on structured light triangulation, time-of-flight method or binocular vision principle.

[0062] S12 mainly relies on the following methods to identify loose parts: first, the RGB data in the RGBD image is used for skin color recognition, and the parts of the human body with skin color are identified as compact parts; secondly, the surface contour of the depth data in the RGBD image is used for judgment. For example, the loose parts of clothes generally have more wrinkles, while the surface contour of the compact parts is relatively smooth and has a certain cylindrical and spherical shape. Based on this feature, a surface contour measurement model is created, and then the parts of the human body that need to be measured are matched and identified, so as to identify the loose parts and compact parts.

[0063] S13 specifically includes: acquiring feature data of a compact part of the human body in the first RGBD image and the second RGBD image in the second posture; and restoring feature parameters of the compact part through an inverse transformation of the deformation according to the feature data.

[0064] The inverse transformation of deformation is to use the regression algorithm in machine learning to fit the relationship before and after deformation, and obtain an empirical formula for inverse transformation processing.

[0065] When a person wears loose clothes, many parts of the body will be covered by the clothes. There are two types of covered parts: close-fitting covering and loose covering. Here, the model of the close-fitting covering (compact) part is assumed to be the corresponding part of the human user model, while the model of the loose covering (loose) part is far from the human user model.

[0066] It is worth noting that different postures and movements can be used to turn non-close-fitting coverage into close-fitting coverage. For example, putting your hands on your hips can turn your waist into a close-fitting coverage, lifting your calves can turn your calves into a close-fitting coverage, raising one hand and bending to the side can turn the side of your chest into a close-fitting coverage, and so on.

[0067] Therefore, by setting postures or movements that can reflect all human body characteristics in advance and letting the tested person perform the postures one by one, the characteristic parameters required by all human user models can be obtained.

[0068] Feature parameters generally include: height, shoulder height, shoulder width, waist circumference, chest circumference, calf and leg contours, arm contours, etc., as well as feature points of each part, i.e., joint points.

[0069] In S15, a standard parameterized human body model is first established; and then the standard parameterized human body model is deformed according to the characteristic parameters to obtain a human body user model.

[0070] S2. Analyze the user's historical purchase records according to user needs and obtain recommended clothing models, including:

[0071] S21. Filter historical purchase records that meet user needs, where the historical purchase records include product evaluation information and return and exchange reason information.

[0072] S22. From the historical purchase records that meet the user's needs, retain the product information whose product evaluation information exceeds the preset target, and at the same time eliminate some product information based on the return reason information to obtain reference product information.

[0073] Clarify user needs, which may include style, color, size, price range, etc.

[0074] Extract product evaluation information from historical purchase records that meet the conditions. Set a preset goal and retain products whose evaluation information meets the preset goal (such as evaluation scores higher than 3.5) from the filtered products.

[0075] Analyze the return reason information to determine the negative reasons that affect product selection (such as size mismatch, quality issues, unsuitable style, etc.). Eliminate specified categories of products based on the return reason. For example, if "unsuitable style" appears frequently in the return reason, then exclude this category of products.

[0076] Finally, we get a list of products with high reviews and low return rates as recommended reference products.

[0077] S23. Analyze the reference product information to make clothing recommendations, and use the corresponding clothing model as the recommended clothing model.

[0078] Various features are extracted from the reference product information, including but not limited to style, color, pattern, material, brand, price range, suitable season, applicable occasion, etc.

[0079] Based on the extracted features, clothing with similar features is searched in the product database.

[0080] In this embodiment, the cosine similarity method is used to calculate the feature similarity, including:

[0081] Convert various features into one-hot encoding representation. For example, style can be represented by a vector, where each element represents a style. If the product belongs to a certain style, the element is 1, otherwise it is 0. Then calculate the cosine similarity between the two product style vectors.

[0082] Finally, the similarities of each feature are weighted and summed to obtain the overall similarity of the two products.

[0083] S3. Determine the recommended clothing model size based on the user body model to obtain the clothing model to be tested.

[0084] The size of the recommended clothing model is determined based on the human body feature parameters in the user's human body model as the clothing model to be tried.

[0085] S4. Performing a fusion operation on the user human body model and the clothing model to be tried on to generate a virtual fitting effect, including:

[0086] S41. Construct a human-clothing fusion network, which includes a first feature extraction network, a second feature extraction network, a feature fusion network, and a transformation network.

[0087] Among them, the first feature extraction network and the second feature extraction network both include a first network layer, a second network layer and a third network layer; the first network layer, the second network layer and the third network layer are connected in sequence, and the first network layer, the second network layer and the third network layer all include a convolutional layer, a normalization layer and a loss function.

[0088] The first feature extraction network obtains human skeleton posture feature data (i.e., skeleton feature map) based on the user human body model, and the second feature extraction network obtains clothing feature data (i.e., clothing feature map) based on the clothing model to be tested. The feature fusion network performs feature fusion processing on the human skeleton posture feature data and the clothing feature data to obtain a preliminary fused feature image and inputs it into the transformation network to obtain a mixed feature image.

[0089] Specifically, the transformation network performs thin plate spline transform (TPS) processing on the preliminary fused feature image to obtain a mixed feature image.

[0090] The calculation expression of the mixed feature image is as follows:

[0091] Corr=A T B

[0092] In the formula, Corr represents mixed correlation features, A represents the skeleton feature map, and B represents the clothing feature map.

[0093] After batch matrix operations, the relevant features are obtained. The dimensions of A are N×C×H×W, the dimensions of B are N×C×W×H, and the dimensions of the mixed feature image are N×C×H×H. Then, the transformed clothing image is obtained according to the TPS grid transformation network. The training loss function L of the human-clothing fusion network is TPS for:

[0094]

[0095] Wherein, E represents the output result of the feature extraction network, K represents the number of TPS transformations, and in this embodiment, K is set to 1-3. E(S) represents the output of the input of the transformation source Source model after passing through the feature extraction network, and E(D) represents the output of the input of the transformation target Dst after passing through the feature extraction network.

[0096] This embodiment describes the thin plate spline transformation (TPS) in the present invention:

[0097] For a given feature map p∈P H×W , where H and W are width and height. For any pixel (x, y) in the image, TPS optimizes the following objectives:

[0098]

[0099] Where R represents the clothing area in the picture. For the image S in area R, represents the thin plate spline transform network;

[0100] The TPS transform is given by the following formula:

[0101] TPS(S,P)=D;(S,D∈P 2 )

[0102] TPS(S, P) represents the TPS transformation formula, D represents the target image after Dst transformation, and S represents the image transformed by Src;

[0103] For the kth TPS transformation of N feature points, the following form is given:

[0104]

[0105] p represents the coordinates of any pixel point in the R region (p = (x, y)), A k Represents the transformation matrix Ak ∈P 2×3 ,ω ki ∈P 2×1 represents the TPS transformation weight, U represents the kernel function, Represents the i-th feature point in image D. The TPS transformation can be approximately represented by k transformations:

[0106]

[0107] TPS(·) represents the thin plate spline transformation function, k represents the kth transformation, p represents the pixel coordinates in the image area, M k (p) represents the contribution mask of point p, S and P represent the pixels in the image area, and T k (p) represents the k-th thin plate spline transformation function;

[0108] The contribution mask of point p is given by:

[0109]

[0110] Represents the predicted output of the model.

[0111] In the thin plate spline transform (TPS), the contribution mask is used to determine the influence of each control point on the final deformation. The contribution mask is a matrix equal to the number of control points, and each control point corresponds to a contribution mask matrix element. For each control point, the value of the contribution mask matrix element represents the degree of influence of the control point on the final deformation. In the thin plate spline transform, the influence of each control point on the final deformation can be adjusted by optimizing the value of the contribution mask matrix element. The optimization process usually uses optimization algorithms such as gradient descent to iteratively update the value of the contribution mask matrix element to minimize the energy function in the deformation process.

[0112] S42. Introduce noise and generate a target dressing effect with a vector representation based on the Prompt semantic description encoder.

[0113] Specifically, a target dressing effect text template is constructed; noise is introduced and added to the target dressing effect text template to obtain the target dressing effect text template after adding noise; the target dressing effect text template after adding noise is input into the Prompt semantic description encoder for text vector transformation processing to obtain the target dressing effect with vector representation.

[0114] S43. Based on the sampling algorithm with Mask, the mixed feature image and the target dressing effect are iteratively sampled through the controlnet control network to obtain the preliminary dressing result.

[0115] The mixed feature image is sampled and processed by the Mask sampling algorithm to obtain sampling data; the sampling data is input into the controlnet control network for optimization and adjustment processing to generate mask data; according to the mask data, similar features of the mixed feature image and the target dressing effect represented by a vector are determined and extracted to obtain the sampling data of the next iteration; based on the sampling data of the next iteration, the above optimization and adjustment processing steps and similar feature extraction steps are looped until the number of iterations meets the preset threshold number, and a preliminary dressing result is output.

[0116] S44. The user body model, the clothing model to be tried on, and the preliminary dressing results are encoded and mixed in sequence through the GAN network to generate a virtual fitting effect.

[0117] The embodiment of the present invention adopts the generation-discrimination loss function mechanism of the adversarial generation network. ξ (x), which distinguishes between real and fake generated mixed images.

[0118] The adversarial loss function is written as follows:

[0119]

[0120] L adv Represents the loss function of the GAN network, L D represents the discriminator loss function, L G represents the generator loss function. Specifically, represents the predicted image, y represents the real image, E y [·]With E x [·] indicates expected value.

[0121] The total loss function L in this method is:

[0122] L=L TPS +L adv .

[0123] On the other hand, the embodiment of the present invention further proposes a personalized virtual fitting recommendation system based on user body shape data, which is used to implement the personalized virtual fitting recommendation method based on user body shape data. Figure 2 As shown, the system includes:

[0124] A human body model building module is used to obtain multi-view human body images of the user and build a human body model of the user;

[0125] The clothing screening module is used to analyze the user's historical purchase records according to user needs and obtain recommended clothing models;

[0126] A clothing model acquisition module to be tested is used to determine the recommended clothing model size based on the user's body model to obtain the clothing model to be tested;

[0127] The fitting effect display module is used to perform a fusion operation on the user's body model and the clothing model to be tried on to generate a virtual fitting effect.

[0128] Preferably, the human body model building module includes:

[0129] A first image acquisition unit, used to acquire a first human body image of a user in a first posture;

[0130] A recognition unit, used to recognize the loose parts of the human body's clothing and the tight parts of the human body in the first human body image;

[0131] A second image acquisition unit is used to guide the user to make at least one predetermined second posture, so that the loose part of the human body clothing becomes a compact part, and acquire at least one second human body image in the second posture of the user;

[0132] A parameter extraction unit, configured to extract human body characteristic parameters according to the first human body image and at least one second human body image;

[0133] The user human body model generating unit is used to create a user human body model according to human body feature parameters.

[0134] Preferably, the clothing screening module includes:

[0135] A preliminary screening unit, used to screen historical purchase records that meet user needs, the historical purchase records include product evaluation information and return reason information;

[0136] A reference information acquisition unit is used to retain the product information whose product evaluation information exceeds the preset target from the historical purchase records that meet the user's needs, and to eliminate part of the product information based on the return reason information to obtain reference product information;

[0137] The clothing recommendation unit is used to analyze the reference product information to make clothing recommendations and use the corresponding clothing models as recommended clothing models.

[0138] Preferably, the fitting effect display module includes:

[0139] A fusion network construction unit is used to construct a human-clothing fusion network, which includes a first feature extraction network, a second feature extraction network, a feature fusion network, and a transformation network; the first feature extraction network obtains human skeleton posture feature data based on a user human body model, the second feature extraction network obtains clothing feature data based on a clothing model to be tested, and the feature fusion network performs feature fusion processing on the human skeleton posture feature data and the clothing feature data to obtain a preliminary fusion feature image and inputs it into the transformation network to obtain a mixed feature image;

[0140] An effect generation unit, used for introducing noise and generating a target dressing effect with a vector representation based on a Prompt semantic description encoder;

[0141] The dressing unit is used to iteratively sample the mixed feature image and the target dressing effect through the controlnet control network based on the sampling algorithm with Mask to obtain the preliminary dressing result;

[0142] The fitting effect display unit is used to encode and mix the user body model, the clothing model to be tried on and the preliminary dressing results in sequence through the GAN network to generate a virtual fitting effect.

[0143] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0144] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A personalized virtual fitting recommendation method based on user body shape data, characterized in that: The following steps are involved: Obtain multi-view human body images of the user and construct a human body model of the user; According to user needs, analyze the user's historical purchase records to obtain recommended clothing models; Determine the recommended clothing model size based on the user body model to obtain a clothing model to be tested; A fusion operation is performed on the user body model and the clothing model to be tried on to generate a virtual fitting effect.

2. The personalized virtual fitting recommendation method based on user body shape data according to claim 1, characterized in that: Obtain multi-view human body images of the user and build a human body model of the user, including: Acquire a first human body image of a user in a first posture; identifying loose parts of human clothing and tight parts of the human body in the first human body image; Instructing the user to make at least one predetermined second posture so that the loose part of the human body clothing becomes a compact part, and obtaining at least one second human body image in the second posture of the user; Extracting human body feature parameters according to the first human body image and at least one second human body image; A user body model is created according to the body feature parameters.

3. The personalized virtual fitting recommendation method based on user body shape data according to claim 1, characterized in that: According to user needs, the user's historical purchase records are analyzed to obtain recommended clothing models, including: Filter historical purchase records that meet user needs, wherein the historical purchase records include product evaluation information and return and exchange reason information; From the historical purchase records that meet the user's needs, retain the product information whose product evaluation information exceeds the preset target, and remove some product information based on the return reason information to obtain reference product information; The reference product information is analyzed to make clothing recommendations, and the corresponding clothing models are used as recommended clothing models.

4. The personalized virtual fitting recommendation method based on user body shape data according to claim 1, characterized in that: Performing a fusion operation on the user body model and the clothing model to be tried on to generate a virtual fitting effect, including: Constructing a human-clothing fusion network, the human-clothing fusion network includes a first feature extraction network, a second feature extraction network, a feature fusion network, and a transformation network; the first feature extraction network obtains human skeleton posture feature data based on a user human body model, the second feature extraction network obtains clothing feature data based on a clothing model to be tested, the feature fusion network performs feature fusion processing on the human skeleton posture feature data and the clothing feature data to obtain a preliminary fusion feature image and inputs the image into the transformation network to obtain a mixed feature image; Introduce noise and generate the target dressing effect with vector representation based on the Prompt semantic description encoder; Based on a sampling algorithm with Mask, the mixed feature image and the target dressing effect are iteratively sampled through the controlnet control network to obtain a preliminary dressing result; The user body model, the clothing model to be tried on and the preliminary dressing result are encoded and mixed in sequence through the GAN network to generate a virtual fitting effect.

5. The personalized virtual fitting recommendation method based on user body shape data according to claim 4, characterized in that: The first feature extraction network and the second feature extraction network both include a first network layer, a second network layer and a third network layer; the first network layer, the second network layer and the third network layer are connected in sequence, and the first network layer, the second network layer and the third network layer both include a convolutional layer, a normalization layer and a loss function.

6. The personalized virtual fitting recommendation method based on user body shape data according to claim 4, characterized in that: The transformation network performs thin plate spline transformation processing on the preliminary fused feature image to obtain a mixed feature image.

7. A personalized virtual fitting recommendation system based on user body shape data, characterized in that: include: A human body model building module is used to obtain multi-view human body images of the user and build a human body model of the user; The clothing screening module is used to analyze the user's historical purchase records according to user needs and obtain recommended clothing models; A clothing model acquisition module to be tested, used to determine the recommended clothing model size based on the user body model to obtain the clothing model to be tested; The fitting effect display module is used to perform a fusion operation on the user body model and the clothing model to be tried on to generate a virtual fitting effect.

8. According to the personalized virtual fitting recommendation system based on user body shape data of claim 7, the human body model building module comprises: A first image acquisition unit, used to acquire a first human body image of a user in a first posture; a recognition unit, configured to recognize a loose part of the human body's clothing and a compact part of the human body in the first human body image; A second image acquisition unit is used to guide the user to make at least one predetermined second posture, so that the loose part of the human body clothing becomes a compact part, and acquire at least one second human body image in the second posture of the user; A parameter extraction unit, configured to extract human body characteristic parameters according to the first human body image and at least one second human body image; The user human body model generating unit is used to create a user human body model according to the human body feature parameters.

9. The personalized virtual fitting recommendation system based on user body shape data according to claim 7, characterized in that: The clothing screening module comprises: A preliminary screening unit, used to screen historical purchase records that meet user needs, wherein the historical purchase records include product evaluation information and return and exchange reason information; A reference information acquisition unit is used to retain the product information whose product evaluation information exceeds the preset target from the historical purchase records that meet the user's needs, and to eliminate part of the product information based on the return reason information to obtain reference product information; The clothing recommendation unit is used to analyze the reference commodity information to make clothing recommendations and use the corresponding clothing model as the recommended clothing model.

10. The personalized virtual fitting recommendation system based on user body shape data according to claim 7, characterized in that: The fitting effect display module comprises: A fusion network construction unit is used to construct a human-clothing fusion network, wherein the human-clothing fusion network includes a first feature extraction network, a second feature extraction network, a feature fusion network, and a transformation network; the first feature extraction network obtains human skeleton posture feature data based on a user human body model, the second feature extraction network obtains clothing feature data based on a clothing model to be tested, the feature fusion network performs feature fusion processing on the human skeleton posture feature data and the clothing feature data to obtain a preliminary fusion feature image and inputs the image into the transformation network to obtain a mixed feature image; An effect generation unit, used for introducing noise and generating a target dressing effect with a vector representation based on a Prompt semantic description encoder; A dressing unit, for iteratively sampling the mixed feature image and the target dressing effect through a controlnet control network based on a sampling algorithm with a mask, to obtain a preliminary dressing result; The fitting effect display unit is used to encode and mix the user body model, the clothing model to be tried on and the preliminary dressing result in sequence through a GAN network to generate a virtual fitting effect.

Images