A method and system for a virtual fitting design service of an underwear

Abstract

The present application relates to the technical field of underwear design service, in particular to a kind of underwear virtual fitting design service method and system;The method comprises obtaining body posture image data and underwear virtual model uploaded by user;After analyzing the smooth area in body posture image data, the smooth parameter of smooth area is confirmed;Based on the smooth parameter, the texture geometric features of wrinkle and depression in body posture image data are recovered using deep neural network, and body posture reconstruction image data is obtained;After extracting the initial sparse feature points in the body posture reconstruction image data, the initial sparse feature points are fitted using surface fitting model, and three-dimensional human body model is obtained;The underwear virtual model is matched with three-dimensional human body model, and model matching parameter is obtained;The underwear virtual model is corrected and designed using model matching parameter, and the corrected underwear parameter is obtained.The purpose of the present application is to solve the problem of inaccurate virtual fitting in the process of virtual fitting of existing underwear.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of underwear design service, in particular to an underwear virtual try-on design service method and system. BACKGROUND

[0002] The existing three-dimensional model of human body shape is mainly obtained by preset training under the premise that the user maintains a standard, symmetrical standing posture and a regular body shape. Since the user's standing posture is not standard during data collection, the soft tissue distribution of each user's body will form subtle and irregular skin surface wrinkles or slight depressions. When collecting user image data, there are areas with blurred edges and low contrast in the image data, which lack clear geometric boundaries and sufficient texture information, making it difficult to be effectively recognized and positioned by traditional feature extraction algorithms. When processing the body wrinkle area with low contrast and blurred edges, these details are mistakenly considered as image noise or unnecessary texture and are subjected to excessive smoothing processing. The processing mechanism analyzes the color and brightness changes of local pixels, and if the changes are small and irregular, they are considered to be noise and are subjected to averaging processing, or are smoothed after edge detection to eliminate jaggedness. Thus, the geometric information essential for subsequent three-dimensional reconstruction is lost. Therefore, it is impossible to establish a reliable correspondence, or the established correspondence has a large error. This results in the generation of slight voids or redundant points in the local three-dimensional model reconstruction process.

[0003] However, when the three-dimensional model has voids, the model will not be able to find an effective contact surface, resulting in the underwear appearing to be floating in the air and not fitting the body, because the simulator cannot detect an effective collision; when there are redundant points, the redundant points may be mistakenly identified as part of the body surface, resulting in unnatural squeezing or insertion of the underwear, especially in critical areas such as the edge of the cup, the connection between the shoulder strap and the skin, and the fitting part of the side wing and the torso. The slight inaccuracy in the critical areas will be magnified, directly affecting the support, shaping and comfort performance of the underwear. Thus, when the user uses the existing virtual interface to see the underwear try-on effect, the virtual underwear may appear to be slightly floating and have unnatural wrinkles in specific body parts such as the edge of the cup, the side wing and the lower back. SUMMARY

[0004] The purpose of the present application is to provide an underwear virtual try-on design service method and system to solve the problem of inaccurate virtual try-on caused by the difficulty of underwear fitting the model body during the virtual try-on process of the existing technology.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is as follows: an underwear virtual try-on design service method, comprising:

[0006] obtaining body posture image data and an underwear virtual model uploaded by a user;

[0007] analyzing the smooth region in the body posture image data, and confirming a smooth parameter of the smooth region;

[0008] based on the smooth parameter, restoring texture geometric features of the wrinkle and the depression in the body posture image data by using a deep neural network to obtain body posture reconstruction image data;

[0009] extracting initial sparse feature points in the body posture reconstruction image data, fitting the initial sparse feature points by using a curved surface fitting model to obtain a three-dimensional human body model;

[0010] matching the underwear virtual model with the three-dimensional human body model to obtain model matching parameters;

[0011] correcting and designing the underwear virtual model by using the model matching parameters to obtain corrected underwear parameters.

[0012] Preferably, after the initial sparse feature points in the body posture reconstruction image data are extracted, the step of fitting the initial sparse feature points by using a curved surface fitting model to obtain a three-dimensional human body model comprises:

[0013] after analyzing the local geometric features in the body posture reconstruction image data, determining deviation information between the local geometric features and preset body shape topological constraints;

[0014] adjusting the preset body shape topological constraints by using the deviation information to obtain specific topological constraints of the user body posture;

[0015] based on the specific topological constraints, after the initial sparse feature points in the body posture reconstruction image data are extracted, fitting the initial sparse feature points by using a curved surface fitting model to obtain a three-dimensional human body model.

[0016] Preferably, after the local geometric features in the body posture reconstruction image data are analyzed, the step of determining deviation information between the local geometric features and preset body shape topological constraints comprises:

[0017] monitoring pixel displacement information and deformation velocity information in the body posture reconstruction image data in a body posture specific region;

[0018] based on the pixel displacement information and the deformation velocity information, obtaining local geometric features in the body posture reconstruction image data;

[0019] based on the local geometric features and the preset body shape topological constraints, obtaining the deviation information.

[0020] Preferably, the step of adjusting the preset body shape topological constraints by using the deviation information to obtain specific topological constraints of the user body posture comprises:

[0021] The degree of deviation between the body shape represented by the deviation information and the preset body shape topological constraints is evaluated.

[0022] Based on the degree of deviation and the preset deviation threshold, after determining the deformation parameter boundary of the preset body topology constraint, the initial specific topology constraint is obtained;

[0023] After performing geometric continuity verification on the initial specific topological constraints, the local discontinuities in the initial specific topological constraints are corrected to obtain the corrected specific topological constraints.

[0024] Preferably, after performing geometric continuity verification on the initial specific topological constraints and correcting the local discontinuities in the initial specific topological constraints to obtain the corrected specific topological constraints, the method further includes:

[0025] Obtain the local curvature variation parameters of the modified specific topological constraint;

[0026] Confirm the degree of matching between the corrected specific topological constraints and the specific body region represented by the deviation information;

[0027] The weights of the energy function in the surface fitting model are determined based on local curvature variation parameters, matching degree, and preset matching threshold.

[0028] Based on the weights of the energy function in the surface fitting model, the modified specific topological constraints are adjusted to obtain the adjusted specific topological constraints.

[0029] Preferably, the step of adjusting the modified specific topological constraints based on the weights of the energy function in the surface fitting model to obtain the adjusted specific topological constraints includes:

[0030] The body-specific regions represented by deviation information are divided, and the local feature stability information of each sub-region is determined.

[0031] Based on the weights of the energy function in the surface fitting model, determine the adjustment step size of the weights;

[0032] After evaluating the local feature stability information of each sub-region, if the evaluation of the local feature stability information is unqualified, the weight of the sub-region corresponding to the local feature stability information is adjusted by adjusting the step size to obtain the weight of each adjusted sub-region.

[0033] Based on the weights of each adjusted sub-region, the modified specific topological constraints are adjusted to obtain the adjusted specific topological constraints.

[0034] Preferably, the step of dividing the body-specific regions represented by the deviation information and determining the local feature stability information of each sub-region includes:

[0035] characterize a posture-specific region based on the bias information, determine a texture density and a curvature variation gradient of the posture-specific region in posture reconstruction image data;

[0036] adjust a sub-region division granularity of the posture-specific region based on the texture density and the curvature variation gradient;

[0037] divide the posture-specific region characterized by the bias information by using the sub-region division granularity of the posture-specific region, and determine local feature stability information of each sub-region.

[0038] Preferably, the step of obtaining the posture image data uploaded by the user and the underwear virtual model comprises:

[0039] obtaining posture image initial data uploaded by the user and an underwear virtual model;

[0040] performing denoising processing on the posture image initial data to obtain denoised posture image initial data;

[0041] extracting a detail region in the denoised posture image initial data, performing local smoothing processing on the detail region, and performing global smoothing processing on a non-detail region in the denoised posture image initial data to obtain posture image data.

[0042] Preferably, the step of analyzing the smooth region in the posture image data and confirming the smooth parameter of the smooth region comprises:

[0043] identifying smooth edge information of the posture image data based on the posture image data;

[0044] analyzing a change region containing pixel gradient changes in the posture image data by using the smooth edge information;

[0045] extracting a smooth parameter of the smooth region after verifying the change region and taking the change region passing the verification as the smooth region.

[0046] The application further provides an underwear virtual fitting design service system, which comprises:

[0047] an obtaining module configured to obtain posture image data uploaded by a user and an underwear virtual model;

[0048] an analyzing module configured to analyze a smooth region in the posture image data and confirm a smooth parameter of the smooth region;

[0049] a restoring module configured to restore texture geometric features of wrinkles and depressions in the posture image data by using a deep neural network based on the smooth parameter, and obtain posture reconstruction image data;

[0050] a fitting module configured to, after extracting the initial sparse feature points in the body posture reconstruction image data, fit the initial sparse feature points using a curved surface fitting model to obtain a three-dimensional human body model;

[0051] a matching module configured to match the underwear virtual model with the three-dimensional human body model to obtain model matching parameters;

[0052] a design module configured to correct the design of the underwear virtual model using the model matching parameters to obtain corrected underwear parameters.

[0053] Compared with the prior art, the underwear virtual try-on design service method and system has the following advantages:

[0054] The underwear virtual try-on design service method and system of the present application obtains the body posture image data and the underwear virtual model uploaded by the user, and finely processes the body posture image data, including analyzing the smooth area and confirming the smooth parameters, and restoring the texture geometric features of the wrinkles and depressions based on the smooth parameters using a deep neural network, to obtain high-precision body posture reconstruction image data. On this basis, the initial sparse feature points in the body posture reconstruction image data are extracted, and a precise three-dimensional human body model is obtained by fitting using a curved surface fitting model. Subsequently, the underwear virtual model is matched with the three-dimensional human body model to obtain model matching parameters, and the underwear virtual model is corrected and designed using these parameters to finally obtain corrected underwear parameters. The restoration of the texture geometric features of the wrinkles and depressions by the deep neural network can make up for the loss of detailed information in the image processing process, so that the reconstructed three-dimensional human body model is closer to the real body posture of the user, especially for users with a plump body shape, the body curves and soft tissue deformation can be more accurately captured. The fit of the underwear with the body is more accurate, thereby significantly improving the realism and accuracy of the virtual try-on. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the specific embodiments of the present application, the drawings required in the specific embodiments will be briefly introduced below. In all the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0056] Figure 1 The flowchart of the underwear virtual try-on design service method of the present application.

[0057] Figure 2 The structural block diagram of the underwear virtual try-on design service system of the present application.

[0058] In the figure: 210, acquisition module; 220, analysis module; 230, recovery module; 240, fitting module; 250, matching module; 260, design module.

[0059] The implementation and advantages of the present application will be further illustrated with reference to the accompanying drawings. DETAILED DESCRIPTION

[0060] The present application will now be described more fully with reference to the accompanying drawings, in which several embodiments of the application will be shown. The present application may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and fully convey the scope of the application to those skilled in the art.

[0061] It should be noted that all directional directions (such as up, down, left, right, front, back, etc.) described herein are only used to explain relative positional relationships, movement conditions, etc. between components, and if the specific posture (as shown in the drawings) changes, the directional directions will also change accordingly.

[0062] In addition, the descriptions such as "first", "second" and the like in the present application are only for the purpose of description, and are not particularly intended to indicate the order or sequence, nor to limit the present application. They are merely used to distinguish components or operations described by the same technical terms, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implying the number of the indicated technical features. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the realization of ordinary skilled in the art. When the combination of technical solutions contradicts each other or cannot be realized, it should be considered that the combination of technical solutions does not exist, nor is it within the scope of protection required by the present application.

[0063] The existing three-dimensional model is usually established when the user maintains a standard symmetrical posture and is trained for a regular body shape. There are limitations for non-standard posture and special body shape. When a user with a plump body shape performs body scanning, due to personal habits and soft tissue distribution characteristics, the user may adopt a non-standard posture slightly deviated to one side or with uneven center of gravity, which is difficult to be effectively recognized and positioned by a traditional feature extraction algorithm. At the same time, the low-contrast and edge-blurred body wrinkle region in the image is mistakenly regarded as image noise or unnecessary texture and is subjected to excessive smoothing processing. Thus, the geometric information essential for subsequent three-dimensional reconstruction is lost. This may further cause the generation of slight holes or redundant points in the three-dimensional model reconstruction process. Subsequently, when simulating the virtual underwear fitting, the parameter set for defining the virtual underwear fabric stretching and bending characteristics, the internal contact mechanics model, when there are inaccurate local body model surfaces with holes or redundant points, the three-dimensional model cannot correctly identify the fitting degree and specific force distribution of the underwear and the body, which directly affects the support, shaping and comfort performance of the underwear. Thus, when the user sees the underwear effect in the virtual fitting interface, the virtual underwear may appear slightly floating and unnatural wrinkles in specific body parts, or even the model may be inserted.

[0064] In order to further understand the content, characteristics and effects of the present application, the following examples are given, and are described in detail as follows with reference to the accompanying drawings:

[0065] Please refer to Figure 1 The present application provides a virtual underwear fitting design service method, comprising the following steps:

[0066] S100, obtaining body posture image data uploaded by a user and a virtual underwear model. The body posture image data is a pre-processed two-dimensional image sequence, which captures image information of the user's body from different angles. The virtual underwear model is a three-dimensional digital representation of underwear, which contains information such as the geometric shape, material properties and physical characteristics of the underwear. Specifically, the body posture image data can be obtained in various ways, for example, the user can use a smartphone or a professional scanning device to take a series of body photos or videos from different angles, which are then uploaded to the service platform. Alternatively, the user can provide existing three-dimensional body scanning data, which is converted into compatible body posture image data format. The virtual underwear model can be provided by the underwear brand or created by a three-dimensional modeling software and uploaded to the system. The virtual underwear model usually contains the geometric shape, material properties and preset physical parameters of the underwear.

[0067] S200, after analyzing the smooth region in the body posture image data, the smooth parameter of the smooth region is confirmed. Wherein, the smooth region is the region with small pixel gradient change and unobvious texture details in the body posture image data, and the smooth region may appear flat due to image processing or body shape. The smooth parameter is a numerical value describing the characteristics of the smooth region, such as the average brightness, color variance or local curvature of the region, which is used for subsequent texture recovery. Specifically, the identification of the smooth region can be realized by image processing algorithm; for example, a method based on pixel gradient can be used, by obtaining the gradient value of each pixel point and its neighborhood in the image, the region with gradient value lower than the preset threshold is identified as the smooth region. Another method is to use image segmentation technology to divide the image into different regions, and then evaluate the texture complexity of each region, and mark the region with low texture complexity as the smooth region. Confirming the smooth parameter can include calculating the average brightness, color variance or local curvature of the smooth region, etc. for subsequent texture recovery.

[0068] S300, based on the smooth parameter, the texture geometric features of the wrinkles and depressions in the body posture image data are recovered by using a deep neural network, and the body posture reconstruction image data is obtained. Wherein, the deep neural network is a machine learning model that learns image features through multiple layers of nonlinear transformation, which can identify and recover hidden geometric information from complex image data. The texture geometric features of the wrinkles and depressions refer to the subtle undulations and depressions formed on the body surface due to skin accumulation, gravity or posture changes, and the texture geometric features are crucial for accurate reconstruction of three-dimensional human body model. The body posture reconstruction image data refers to the image data that has recovered the texture geometric features of wrinkles and depressions after being processed by the deep neural network, which contains more detailed information than the original body posture image data. Specifically, the training of the deep neural network requires a large amount of real body image data containing wrinkles and depression details and its corresponding three-dimensional model data. And the smooth parameter can be used as one of the inputs of the deep neural network to guide the network to pay attention to and recover the over-smoothed region. For example, a generative adversarial network (GAN) architecture can be used, in which the generator is responsible for recovering the texture details from the smooth region, and the discriminator evaluates the authenticity of the recovery result. In this way, the deep neural network can learn the internal geometric laws of wrinkles and depressions, so as to reconstruct these key texture geometric features in the body posture image data and generate body posture reconstruction image data containing more details.

[0069] S400, after extracting the initial sparse feature points in the body posture reconstruction image data, fitting the initial sparse feature points by using a curved surface fitting model to obtain a three-dimensional human body model. Wherein, the initial sparse feature points refer to the key points extracted from the body posture reconstruction image data for constructing a three-dimensional model, which are usually distributed at the skeletal landmarks or significant geometric features of the body. The curved surface fitting model is a mathematical model used to fit discrete feature points into continuous three-dimensional surfaces, thereby constructing a complete three-dimensional human body model. The three-dimensional human body model refers to the three-dimensional digital representation of the user's body, which can accurately reflect the user's body shape, posture and surface details. Specifically, the extraction of initial sparse feature points can use classical computer vision algorithms such as Scale-Invariant Feature Transform (SIFT) or Speeded Up Robust Features (SURF), so as to identify key points with good stability and repeatability in images. Alternatively, a pre-trained human pose estimation model is used to directly identify human skeletal key points or anatomical landmarks from the body posture reconstruction image data as initial sparse feature points. After obtaining the initial sparse feature points, various curved surface fitting models can be used to construct the three-dimensional human body model. For example, methods based on spline curves or B-spline surfaces can be used to construct smooth body surfaces by interpolating or approximating these feature points. Another method is to use a statistical shape model (such as the SMPL model) to match the initial sparse feature points by adjusting the parameters of the model, thereby obtaining a three-dimensional human body model that conforms to the user's body shape.

[0070] S500, matching the underwear virtual model with the three-dimensional human body model to obtain model matching parameters. Wherein, the model matching parameters refer to the geometric transformation parameters obtained after matching the underwear virtual model with the three-dimensional human body model, such as translation, rotation and scaling, as well as the fitting degree of the underwear on the body and the stress distribution information. Specifically, the matching process determines the optimal position and posture of the underwear on the user's body. This is achieved through geometric transformation algorithms, for example, by minimizing the distance or overlapping area between the underwear model and the body model to determine the optimal translation, rotation and scaling parameters. Or use a physics simulation engine to place the underwear virtual model on the three-dimensional human body model and simulate physical effects such as gravity and friction to observe the natural sagging and fitting of the underwear on the body, thereby obtaining more realistic model matching parameters. The model matching parameters not only include geometric transformation information, but also include physical information such as pressure distribution and stretching degree of the contact area between the underwear and the body.

[0071] S600, correcting the underwear virtual model according to the model matching parameters to obtain corrected underwear parameters. The corrected underwear parameters refer to the parameters obtained by correcting the underwear virtual model according to the model matching parameters. The corrected underwear parameters can be used to guide the production of actual underwear or further virtual fitting optimization. Specifically, the correction can be performed according to the fit of the underwear and the body reflected in the model matching parameters. For example, if the model matching parameters show that the underwear has floating or squeezing in a certain area, the size, shape or cutting of the underwear virtual model can be adjusted to improve the fit. The adjustment can be realized by a parameterized design tool, or the parts of the underwear model can be manually adjusted. Another method is to use an optimization algorithm to automatically adjust the parameters of the underwear virtual model according to the preset target functions such as comfort and aesthetics, until the best fit is achieved, thereby obtaining the corrected underwear parameters. The corrected parameters can be used to guide the production of actual underwear, or as the basis for user's personalized customization.

[0072] In this embodiment, by obtaining the body image data uploaded by the user and the underwear virtual model, a foundation is laid for subsequent analysis and design. By analyzing the smooth areas in the body image data and confirming the smooth parameters, the areas in the image that may have lost details due to excessive smoothing can be identified, providing guidance for subsequent texture restoration. For example, in areas with rich soft tissue such as the side waist or the lower back, due to the smoothing algorithm of the image signal processor, the texture information of wrinkles and depressions may be blurred. By identifying these smooth areas and extracting their smooth parameters, the texture geometry features of wrinkles and depressions in the body image data are restored using a deep neural network based on the smooth parameters, obtaining body reconstruction image data. The deep neural network can learn and reconstruct the body details that have been erased by traditional image processing algorithms, such as subtle skin wrinkles and depressions. For example, when the user adopts a non-standard standing posture, irregular skin accumulation or depressions may occur in certain parts of the body, which may be unclear in the original image. Through learning a large amount of real body data, the deep neural network can infer and restore these complex texture geometry features from the smooth areas, thereby generating body reconstruction image data that is closer to the real body shape. Avoiding the inaccuracy of the three-dimensional model due to the lack of image details. Subsequently, initial sparse feature points are extracted from the body reconstruction image data, and these feature points are fitted using a surface fitting model, ultimately obtaining a high-precision three-dimensional human body model. Since the body reconstruction image data contains rich texture geometry features, the initial sparse feature points extracted will be more accurate and dense, better capturing the complex curves of the body. The surface fitting model can then construct the discrete feature points into a continuous, smooth three-dimensional surface that conforms to the anatomical features of the human body. For example, in the body reconstruction image data that has restored the details of the side waist wrinkles, more feature points reflecting the shape of the wrinkles can be extracted, enabling the final generated three-dimensional human body model to accurately present the concave-convex morphology of the wrinkles. Finally, the underwear virtual model is matched with the high-precision three-dimensional human body model to obtain model matching parameters, and the underwear virtual model is modified and designed using the model matching parameters to obtain the modified underwear parameters. Since the three-dimensional human body model accurately reflects the user's real body shape and details, the matching of the underwear virtual model with the body model will be more accurate. The model matching parameters not only include the geometric position of the underwear on the body, but also reflect the degree of fit and force distribution of the underwear with the body, etc. For example, if the three-dimensional human body model accurately presents the real shape of the user's chest or hips, the underwear virtual model can more accurately fit these areas when matching, avoiding floating or penetrating. Based on these accurate matching parameters, targeted modification and design of the underwear virtual model can be performed, such as adjusting the shape of the cup, the length of the shoulder strap, or the width of the side wing, to ensure that the underwear presents the best fit and comfort in the virtual fitting, ultimately obtaining the modified underwear parameters that meet the user's individual needs. Thus, the realism and accuracy of virtual fitting are significantly improved.

[0073] In some embodiments of the present application, after extracting the initial sparse feature points in the body posture reconstruction image data, the step of fitting the initial sparse feature points using a curved surface fitting model to obtain a three-dimensional human body model includes:

[0074] After analyzing the local geometric features in the body posture reconstruction image data, the deviation information between the local geometric features and the preset body shape topology constraint is determined. Specifically, this step refers to identifying and quantifying the geometric properties in the body posture reconstruction image data that reflect the local details of the human body surface shape, curvature, and texture, etc. For example, the curvature, gradient, and edge information of each part in the body posture reconstruction image data can be extracted as local geometric features. The preset body shape topology constraint usually refers to the structure and connection relationship of the predefined three-dimensional human body model, such as the topology structure of a standard or average human body model, to provide a basic framework for the construction of the three-dimensional model. The deviation information refers to the difference between the local geometric features and the preset body shape topology constraint, which quantifies the degree of deviation between the actual body posture of the user and the standard body shape.

[0075] The preset body shape topology constraint is adjusted using the deviation information to obtain a specific topology constraint of the user's body posture. This step is to deform the preset general topology constraint locally or globally according to the obtained deviation information, so that it is more consistent with the actual body posture characteristics of the user. For example, if the deviation information indicates that there is a significant difference between the curvature of the user's waist area and the preset topology constraint, the waist area of the preset topology constraint is adjusted accordingly. Thus, after adjustment, the topology structure that more accurately reflects the individual body posture characteristics of the user is obtained, thereby obtaining a specific topology constraint of the user's body posture, which aims to provide more accurate guidance for subsequent three-dimensional model fitting.

[0076] Based on the specific topology constraint, after extracting the initial sparse feature points in the body posture reconstruction image data, the initial sparse feature points are fitted using a curved surface fitting model to obtain a three-dimensional human body model. The specific topology constraint is used to constrain the curved surface fitting model. When fitting the initial sparse feature points, the curved surface fitting model can generate a three-dimensional human body model that conforms to the actual body posture of the user and has a reasonable topology structure under the guidance of the specific topology constraint.

[0077] In this embodiment, by analyzing the local geometric features in the body posture reconstruction image data and comparing them with the preset body shape topological constraints, deviation information is obtained. The deviation information is used to dynamically adjust the preset general body shape topological constraints, so that they are converted into specific topological constraints that are highly matched with the actual body posture of the user. When the initial sparse feature points are extracted and fitted using the surface fitting model in the subsequent process, the surface fitting model is no longer blindly fitted, but is precisely guided by the specific topological constraints. This makes the generated body posture three-dimensional human body model not only accurately capture the detailed features of the user's body posture, but also ensure the rationality and consistency of the overall topological structure, avoiding model distortion or inaccuracy caused by relying solely on sparse feature point fitting. Thus, the fidelity of the three-dimensional human body model is significantly improved; it can ensure a higher matching degree between the virtual underwear and the user's body posture, thereby providing a more realistic try-on experience and more accurate design correction suggestions.

[0078] In some embodiments of the present application described above, after analyzing the local geometric features in the body posture reconstruction image data, the step of determining the deviation information between the local geometric features and the preset body shape topological constraints includes:

[0079] The pixel displacement information and deformation velocity information are obtained by monitoring the pixel displacement and deformation velocity in the body posture specific region in the body posture reconstruction image data. This step aims to capture the pixel position changes and deformation rates caused by subtle changes in body posture in the image. The body posture specific region can be understood as the region of the human body that is prone to wrinkles, depressions or deformation when wearing underwear, such as the chest, waist and hips, etc. The pixel displacement information and deformation velocity information are indicators that quantify the dynamic changes in these regions. Through image processing techniques such as optical flow method, feature point tracking, etc., these information can be accurately obtained.

[0080] Based on the pixel displacement information and deformation velocity information, the local geometric features in the body posture reconstruction image data are obtained. The local geometric features are the shape, curvature and texture details of the body posture in the specific region. For example, by analyzing the pattern of pixel displacement and the size of deformation velocity, the geometric properties of the region such as convexity, concavity and surface smoothness can be inferred.

[0081] Based on the local geometric features and the preset body shape topological constraints, the deviation information is obtained. The preset body shape topological constraints are standard idealized three-dimensional human body models that contain the topological structure and geometric relationship of each part of the human body. The deviation information is the difference between the local geometric features of the actual body posture and the preset body shape topological constraints. This difference can be a deviation in shape, a mismatch in size, or an inconsistency in local details. Thus, the specific deviation between the user's body posture and the standard model can be quantified.

[0082] In this embodiment, by monitoring the pixel displacement and deformation velocity of the body state specific region in the body state reconstruction image data, the dynamic changes and subtle deformations of the body state can be captured, thereby obtaining more fine and accurate local geometric features. By introducing pixel displacement information and deformation velocity information, the application can more comprehensively represent the local morphology of the body state, providing a more reliable basis for subsequent comparison with the preset body type topological constraint. Due to the ability to obtain high-precision local geometric features, the real deviation of the body state can be more accurately identified when compared with the preset body type topological constraint, thereby providing accurate input for subsequent three-dimensional human body model fitting and underwear correction design. Moreover, by monitoring the pixel displacement and deformation velocity, the dynamic changes and deformation trends of the body state in subtle places can be captured, thereby accurately identifying complex texture geometric features such as wrinkles and depressions. The obtained deviation information is more accurate and reliable. Accurate deviation information has a significant positive effect on subsequent construction of high-fidelity three-dimensional human body models using surface fitting models and personalized underwear correction design.

[0083] In some embodiments of the application described above, the step of adjusting the preset body type topological constraint using the deviation information to obtain a specific topological constraint of the user's body state includes:

[0084] Evaluating the deviation degree of the body state represented by the deviation information from the preset body type topological constraint. This step refers to calculating or analyzing the difference in geometric shape between the user's body state and the preset body type topological constraint reflected by the deviation information. The deviation degree can be measured in various ways, for example, the Euclidean distance between corresponding points, the angle between surface normal vectors, or the difference based on local curvature, etc. The purpose is to quantify the difference between the user's actual body state and the general body type template, providing a basis for subsequent adjustment.

[0085] After determining the deformation parameter boundary of the preset body type topological constraint based on the deviation degree and the preset deviation threshold, an initial specific topological constraint is obtained. After quantifying the deviation degree, this step determines which regions have significant deviation that need to be adjusted greatly, and which regions have deviation within an acceptable range according to the preset deviation threshold. The determination of the deformation parameter boundary aims to limit the deformation range of the preset body type topological constraint during the adjustment process to prevent model distortion caused by excessive deformation. For example, a maximum displacement distance or a maximum curvature change rate can be set for each vertex or local region. By deforming within these boundaries, an initial specific topological constraint that conforms to the user's body state characteristics can be obtained.

[0086] After the geometric continuity check on the initial specific topological constraint, the local discontinuous region in the initial specific topological constraint is modified to obtain the modified specific topological constraint. In this step, the geometric continuity check refers to checking whether the surface of the initial specific topological constraint is smooth, whether there are sharp corners, cracks or self-intersections, and other geometric defects that do not conform to the natural shape of the human body. For example, curvature analysis and normal vector smoothness check methods can be used for checking. If a local discontinuous region is found, it needs to be modified, for example, by using local smoothing algorithms, spline interpolation or mesh optimization techniques to eliminate these discontinuities, to ensure that the final modified specific topological constraint is continuous and smooth in geometry, and more truly reflects the user's body shape.

[0087] Specifically, after analyzing the local geometric features in the body shape reconstruction image data, the deviation information between the local geometric features and the preset body shape topological constraint is obtained. In order to adjust the preset body shape topological constraint using the deviation information, first, the Euclidean distance between each pair of corresponding vertex pairs in the deviation information can be calculated to evaluate the degree of deviation of the body shape represented by the deviation information from the preset body shape topological constraint. For example, if the Euclidean distance of a certain region exceeds a preset deviation threshold, such as 5 mm, it is considered that there is a significant deviation in this region. Based on this degree of deviation, a deformation parameter boundary can be determined for each vertex of the preset body shape topological constraint. For example, for a region with significant deviation, the vertex is allowed to have a maximum displacement of 10 mm in the normal direction; for a region with insignificant deviation, the maximum displacement is limited to 2 mm. Within this deformation parameter boundary, the preset body shape topological constraint is deformed by minimizing an energy function (for example, an energy function containing a data term and a regularization term), thereby obtaining an initial specific topological constraint. Subsequently, a geometric continuity check is performed on the initial specific topological constraint. Specifically, the average curvature of each vertex on the surface of the initial specific topological constraint can be calculated, and it is checked whether there is a curvature jump or an abnormally large curvature value. If a curvature value exceeding a preset threshold, such as a curvature value representing a sharp edge, is found in a local region, it is considered that there is a local discontinuity in this region. At this time, a local Laplacian smoothing algorithm or a Catmull-Clark subdivision surface algorithm can be used to modify the local discontinuous region until the curvature variation of the region tends to be gentle, thereby obtaining a modified specific topological constraint. In this way, it is ensured that the final specific topological constraint not only fits the user's body shape, but also has good continuity and smoothness in geometry. Thus, by quantifying the degree of deviation and setting the deformation parameter boundary, the accuracy and reasonableness of the adjustment are ensured; by geometric continuity check and modification, the geometric quality and realism of the specific topological constraint are significantly improved. The generated three-dimensional human body model can more accurately capture the body shape features of the user, providing a more reliable and detailed base model for subsequent virtual underwear fitting design, thereby improving the realism of virtual fitting and user experience.

[0088] In some embodiments of the above-mentioned embodiments of the present application, after the step of performing geometric continuity verification on the initial specific topological constraint and modifying the local discontinuous region in the initial specific topological constraint to obtain a modified specific topological constraint, the method further comprises:

[0089] Obtaining a local curvature variation parameter of the modified specific topological constraint. This step refers to determining the curvature variation rate of the modified specific topological constraint in different regions by analyzing the geometric shape of the modified specific topological constraint. The local curvature variation parameter can reflect the bending degree and geometric complexity of the topological constraint in the local region. For example, a high curvature value may indicate a joint, a recess or a protrusion, etc. detailed feature, while a low curvature value may correspond to a relatively flat region. The purpose is to quantify the local geometric characteristics of the topological constraint, and provide a basis for subsequent matching degree evaluation and weight determination.

[0090] Confirming the matching degree of the modified specific topological constraint and the body state specific region represented by the deviation information. Wherein, this step can be understood as evaluating to what extent the modified specific topological constraint matches the actual body state feature region indicated by the deviation information in the body state reconstruction image data. The matching degree can be quantified by calculating the geometric distance, shape similarity or feature point correspondence between the topological constraint and the body state specific region. The purpose is to ensure that the modified topological constraint is not only continuous itself, but also can accurately fit the real geometric shape of the user's body state.

[0091] Based on the local curvature variation parameter, the matching degree and the preset matching threshold, the weight of the energy function in the surface fitting model is determined. Wherein, this step specifically refers to dynamically assigning weights to different energy terms (such as data terms and smoothing terms, etc.) in the surface fitting model according to the geometric complexity reflected by the local curvature variation parameter and the fitting degree reflected by the matching degree, combined with the preset matching threshold. For example, in the region with large curvature variation and low matching degree, the weight of the data term can be appropriately increased to emphasize the fidelity to the original body state feature; while in the region with small curvature variation and high matching degree, the weight of the smoothing term can be increased to ensure the overall smoothness of the model. The purpose is to provide fine guidance for the subsequent surface fitting process, so that the fitting result can better balance the data fidelity and the model smoothness.

[0092] Adjust the modified specific topological constraint based on the weight of the energy function in the surface fitting model to obtain an adjusted specific topological constraint. This step refers to using the above-mentioned energy function weight to further fine-tune the modified specific topological constraint through an iterative optimization algorithm or deformation technology. The adjustment process aims to make the topological constraint better adapt to the optimization objective of the surface fitting model while meeting the geometric continuity, so as to obtain an adjusted specific topological constraint that can accurately capture the body posture details and has good geometric characteristics.

[0093] In this embodiment, the weight of the energy function in the surface fitting model is dynamically determined based on these evaluation results and a preset matching threshold, which realizes fine control of the surface fitting process. For example, in areas with rich body posture details and dramatic curvature changes, higher weights can be given to data items to ensure that the fitting model can faithfully reproduce these details; while in relatively flat areas, the weight of the smoothing term can be increased to avoid introducing unnecessary noise. Finally, by using these adjusted energy function weights to further adjust the modified specific topological constraint, the topological constraint is not only geometrically continuous, but also semantically and geometrically more accurate to the user's body posture, laying a solid foundation for subsequent generation of high-precision three-dimensional human body models. And by obtaining the local curvature change parameter and confirming the matching degree, the geometric characteristics of the topological constraint and its fit with the actual body posture can be more comprehensively understood, thereby avoiding detail distortion or excessive smoothing that may be caused by only focusing on geometric continuity. Based on the dynamic determination of the weight of the energy function in the surface fitting model, the surface fitting process can be adaptively adjusted according to the characteristics of different regions of the body posture, achieving the best balance between data fidelity and model smoothness. Finally, by fine-tuning the modified specific topological constraint, the adjusted specific topological constraint can more accurately capture the subtle features and complex geometric structures of the user's body posture, thereby generating a high-precision, high-realism three-dimensional human body model.

[0094] In some embodiments of the application described above, the step of adjusting the modified specific topological constraint based on the weight of the energy function in the surface fitting model to obtain an adjusted specific topological constraint comprises:

[0095] The body posture specific region represented by the deviation information is divided, and the local feature stability information of each sub-region is determined. Specifically, this step refers to subdividing the body posture region indicated by the deviation information that needs to be adjusted by the topological constraint into multiple smaller sub-regions according to its geometric characteristics or stability requirements. For each sub-region, the stability of its local geometric features, such as the uniformity of the texture and the smoothness of the curvature change, is further analyzed to quantify the stability of the sub-region in the topological adjustment process. The purpose is to provide regional evaluation basis for subsequent fine weight adjustment.

[0096] Based on the weights of the energy function in the surface fitting model, the adjustment step of the weights is determined. This step can be understood as setting the increment or decrement for fine-tuning the weights according to the currently determined energy function weights. The step can be dynamically determined according to the size of the overall weight and the requirement for adjustment precision or the preset adjustment strategy, to ensure the effectiveness and convergence of weight adjustment. The purpose is to provide controllable quantization parameters for subsequent local weight adjustment.

[0097] After evaluating the local feature stability information of each sub-region, if the local feature stability information evaluation is unqualified, the weight of the sub-region corresponding to the local feature stability information is adjusted using the adjustment step, to obtain each adjusted sub-region weight. Specifically, after quantitatively evaluating the local feature stability of each sub-region, if it is found that the stability of a certain sub-region does not meet the preset standard, such as too low or too high stability, indicating that the topological constraint of this region may have problems or needs more fine adjustment, the weight of this sub-region in the energy function of the surface fitting model will be locally adjusted according to the pre-determined adjustment step. In this way, the influence of a specific sub-region in the fitting process can be targetedly enhanced or weakened, to better adapt to its local geometric characteristics. The purpose is to realize adaptive weight adjustment of different regions and improve the local precision of fitting.

[0098] Based on each adjusted sub-region weight, the modified specific topological constraint is adjusted to obtain an adjusted specific topological constraint. Specifically, after locally optimizing the weights of all sub-regions that need to be adjusted, these adjusted sub-region weights are applied to the surface fitting model, so as to finally adjust the overall modified specific topological constraint. Through this fine-grained weight distribution and adjustment, it can be ensured that the three-dimensional human body model can obtain the best fitting effect in different regions, so as to obtain a more accurate and realistic adjusted specific topological constraint.

[0099] Specifically, in generating the three-dimensional human body model, there are multiple body posture specific regions in the posture reconstruction image data, such as shoulder, waist and hip, and the geometric characteristics and stability of these regions are different. First, the posture specific regions indicated by the deviation information are divided, for example, the shoulder region is divided into scapula sub-region and clavicle sub-region, the waist region is divided into side waist sub-region and abdominal sub-region, and the local feature stability information of each sub-region is determined respectively. For example, the scapula sub-region may have low local feature stability due to muscle activity, while the abdominal sub-region may be relatively smooth and have high local feature stability. Then, based on the weight of the energy function in the surface fitting model, an adjustment step of the weight is determined, for example, set to 0.05. Subsequently, the local feature stability information of each sub-region is evaluated. If the local feature stability information of the scapula sub-region is not qualified (for example, its stability is lower than a preset threshold, indicating that the fitting of this region may have a large deviation), the weight of the scapula sub-region is adjusted by the adjustment step of 0.05, for example, the weight is increased by 0.05, to enhance the influence of this region in the fitting process, so that it is more closely fitted to the original data. On the contrary, if the local feature stability information of the abdominal sub-region is qualified, its weight may remain unchanged or be fine-tuned. Finally, based on the adjusted weight of each sub-region, the system will re-adjust the corrected specific topological constraints. By fine-tuning the adaptive weight adjustment for different sub-regions, it can ensure that the details of complex regions such as shoulders are more accurately preserved and fitted, while smooth regions maintain their smoothness, so that the adjusted specific topological constraints are more accurate and realistic overall and conform to the user's body posture characteristics. Thus, it can effectively avoid overfitting or underfitting in complex or unstable regions, significantly improving the geometric accuracy and realism of the three-dimensional human body model in details such as wrinkles and depressions.

[0100] In some embodiments of the present application described above, the step of dividing the posture specific regions represented by the deviation information and determining the local feature stability information of each sub-region includes:

[0101] Based on the posture specific regions represented by the deviation information, the texture density and curvature change gradient of the posture specific regions in the posture reconstruction image data are determined. The texture density can be understood as the frequency and distribution of changes in pixel gray value or color value in the region, reflecting the richness of details in the region. The curvature change gradient represents the bending degree and its change rate of the surface geometry of the region, reflecting the smoothness or ruggedness of the region. It can be calculated by image processing algorithms, such as methods based on gradient, Hessian matrix or local binary pattern (LBP). The purpose is to obtain the fine geometric and texture information of the posture specific regions, providing a data basis for subsequent region division.

[0102] Based on the texture density and the curvature variation gradient, the sub-region division granularity of the posture-specific region is adjusted. Among them, this step is to dynamically adjust the fineness when the posture-specific region is divided according to the determined texture density and curvature variation gradient. For example, for a region with high texture density and large curvature variation gradient, such as a joint or a wrinkle, a smaller division granularity can be used, that is, the region is divided into more smaller sub-regions, so as to capture the local features more finely; and for a region with low texture density and small curvature variation gradient, such as a flat torso part, a larger division granularity can be used, that is, the region is divided into fewer larger sub-regions, so as to improve the processing efficiency.

[0103] The posture-specific region represented by the deviation information is divided by using the sub-region division granularity of the posture-specific region, and the local feature stability information of each sub-region is determined. Specifically, the division process can use algorithms such as quadtree decomposition, K-means clustering, or graph-based segmentation. After the division is completed, the local feature stability information of each sub-region needs to be evaluated. The local feature stability information can include indicators such as the variance of the pixel values in the sub-region, the consistency of the gradient direction, or the uniformity of the curvature. The stability and reliability of the features in the sub-region are measured, for example, a sub-region with small variance and high gradient direction consistency is generally considered to be a feature stable region, and vice versa. The purpose is to provide a basis for the adjustment of the energy function weight in the subsequent process, and to ensure that a greater weight is applied in the feature unstable region to perform more fine adjustment.

[0104] In this embodiment, by determining the texture density and the curvature variation gradient of the posture-specific region, the local details and geometric shapes of the region can be comprehensively understood. Based on these fine local features, the division granularity of the sub-region is adaptively adjusted, so that the division result can better reflect the actual complexity of the posture. For example, in a region with dramatic posture changes or rich details, the division granularity is finer, which can more accurately capture the local features; and in a relatively flat posture region, the division granularity is coarser, which can avoid unnecessary computational overhead. Therefore, the local feature stability information of each sub-region can be more accurately evaluated, providing a more reliable basis for the determination and adjustment of the energy function weight in the subsequent surface fitting model. Through the fine and adaptive region division and feature stability information evaluation mechanism, it is ensured that in the subsequent topological constraint adjustment process, different regions can be processed differently according to their feature differences, thereby improving the fitting accuracy and realism of the three-dimensional human body model.

[0105] In some embodiments of the application described above, the step of obtaining the user-uploaded posture image data and the virtual underwear model includes:

[0106] Obtain the body posture image initial data uploaded by the user and the underwear virtual model. Among them, this step refers to receiving the original body posture image uploaded by the user through the client, which may not have been pre-processed, and receiving the underwear virtual model selected or uploaded by the user. Among them, the body posture image initial data can be understood as the original image data directly obtained from the camera or scanning device, which may contain various environmental noise or acquisition errors.

[0107] Perform denoising processing on the body posture image initial data to obtain denoised body posture image initial data. The purpose of this step is to eliminate random noise, artifacts or unnecessary interference in the image, so as to improve the clarity of the image and the accuracy of subsequent processing. In actual application, denoising processing can use various algorithms, such as Gaussian filtering, median filtering, bilateral filtering or denoising network based on deep learning, to adapt to different noise types and image characteristics.

[0108] After extracting the detail area in the denoised body posture image initial data, performing local smoothing processing on the detail area, and performing global smoothing processing on the non-detail area in the denoised body posture image initial data, obtaining the body posture image data. Among them, the detail area usually refers to the area in the image that contains rich texture, edge or high-frequency information, such as human joints, muscle lines and skin wrinkles. The non-detail area usually refers to the area in the image that is relatively flat and has small texture changes, such as the surface of the trunk or limbs. Local smoothing processing refers to smoothing the small noise or irregularities inside the detail area while maintaining its important features, so as to avoid excessive blurring. Global smoothing processing refers to more extensive smoothing of non-detail areas to eliminate low-frequency noise and make the pixel values of these areas more uniform, thereby providing more stable input for subsequent analysis.

[0109] In this embodiment, denoising processing can effectively remove random interference in the image and lay the foundation for subsequent analysis. Secondly, by distinguishing between detail areas and non-detail areas and applying local smoothing processing and global smoothing processing respectively, it is ensured that key body posture details in the image are retained and optimized, and the uniformity of the smoothed area is also improved. Avoiding the loss of details or insufficient smoothing that may be caused by traditional single smoothing methods, a high-quality body posture image data is obtained. Therefore, the present application can significantly improve the accuracy and realism of the three-dimensional human body model construction, thereby optimizing the matching effect of the underwear virtual model and the three-dimensional human body model, and finally providing an underwear correction design scheme that is more in line with the user's body posture, improving the overall accuracy and user experience of the virtual try-on design service.

[0110] In some embodiments of the present application described above, after analyzing the smoothed area in the body posture image data, the step of confirming the smoothing parameter of the smoothed area includes:

[0111] Based on the body posture image data, smooth edge information of the body posture image data is identified. Specifically, identifying the smooth edge information of the body posture image data refers to detecting the boundary where the pixel intensity or color changes significantly in the body posture image data through an image processing algorithm, such as an edge detection operator (e.g., Canny, Sobel, or Laplacian operator). The edge information helps to distinguish different regions in the image, especially the boundary between smooth and non-smooth regions.

[0112] Using the smooth edge information, a change region containing pixel gradient changes in the body posture image data is analyzed. This step is based on the identification of the smooth edge information and further analyzes the pixel gradient in the image. The pixel gradient changes reflect the texture and detail richness of the local region of the image. The change region refers to a region where the pixel gradient changes are more obvious, which may include wrinkles, depressions, or clothing edges. By analyzing these change regions, potential non-smooth regions can be preliminarily screened out.

[0113] After verifying the change region and confirming that the verified change region is a smooth region, the smooth parameters of the smooth region are extracted. Specifically, this step refers to further evaluating and verifying the preliminarily identified change region. The verification process can include analyzing the local variance, texture complexity, or comparison with the preset smoothness threshold in the region. Only by passing the verification and confirming that it meets the definition of a smooth region, such as the pixel gradient change being below a certain threshold or the texture feature being not obvious, the region is finally determined as a smooth region. Once the smooth region is confirmed, its smooth parameters can be extracted, such as the average pixel value, standard deviation, or specific smoothness indicators for subsequent deep neural network processing. These parameters will guide the subsequent texture and geometric feature restoration process.

[0114] In this embodiment, by identifying the smooth edge information in the body posture image data, the boundaries of different regions in the image can be effectively defined, providing a basis for subsequent regional analysis. By analyzing the change region of the pixel gradient change using the smooth edge information, the regions in the image that may contain complex textures or non-smooth features are preliminarily screened out. Subsequently, through strict verification of these change regions, only regions that truly meet the definition of smooth regions are confirmed as smooth regions. This can avoid misjudging regions containing subtle wrinkles or depressions as smooth regions, thereby ensuring the accuracy of the extracted smooth parameters. Accurate smooth parameters are the key input for subsequent deep neural network restoration of wrinkle and depression texture and geometric features, directly affecting the quality of the reconstructed body posture image data.

[0115] Based on any one of the underwear virtual try-on design service methods in the above embodiments, please refer to Figure 2The application also provides an underwear virtual fitting design service system, which comprises an acquisition module 210, an analysis module 220, a recovery module 230, a fitting module 240, a matching module 250 and a design module 260.

[0116] The acquisition module 210 is used for acquiring body image data uploaded by a user and an underwear virtual model.

[0117] The analysis module 220 is used for confirming a smooth parameter of a smooth area in the body image data after analyzing the smooth area.

[0118] The recovery module 230 is used for recovering texture geometric features of wrinkles and depressions in the body image data by using a deep neural network based on the smooth parameter, so as to obtain body reconstruction image data.

[0119] The fitting module 240 is used for fitting initial sparse feature points in the body reconstruction image data by using a curved surface fitting model after extracting the initial sparse feature points, so as to obtain a three-dimensional human body model.

[0120] The matching module 250 is used for matching the underwear virtual model with the three-dimensional human body model, so as to obtain model matching parameters.

[0121] The design module 260 is used for correcting the design of the underwear virtual model by using the model matching parameters, so as to obtain corrected underwear parameters.

[0122] In this embodiment, the acquisition module 210 is used for acquiring body image data uploaded by a user and an underwear virtual model, which lays a foundation for subsequent analysis and design. The analysis module 220 is used for analyzing a smooth area in the body image data and confirming a smooth parameter. Thus, the area in the image that may lose details due to excessive smoothing can be identified, which provides guidance for subsequent texture recovery. For example, in the soft tissue-rich areas such as the side waist or the lower back, the texture information of wrinkles and depressions may be blurred due to the smoothing algorithm of the image signal processor. The recovery module 230 can infer and recover these complex texture geometric features from the smooth area by learning a large amount of real body data, so as to generate body reconstruction image data that is closer to the real body shape. The fitting module 240 extracts initial sparse feature points from the body reconstruction image data and fits these feature points by using a curved surface fitting model, so as to finally obtain a high-precision three-dimensional human body model. The matching module 250 matches the underwear virtual model with the high-precision three-dimensional human body model, so as to obtain model matching parameters. The design module 260 corrects the design of the underwear virtual model by using these parameters, so as to obtain corrected underwear parameters. Furthermore, through accurate matching and correction design, the edge of the cup of the underwear is ensured to be closely fitted to the body, so as to provide a more reliable virtual fitting experience and personalized underwear design suggestion for the user.

[0123] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some or all of the technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the present application.

Claims

1. A method for providing virtual try-on design services for underwear, characterized in that, include: Obtain user-uploaded body image data and virtual underwear models; After analyzing the smoothed regions in the body image data, the smoothing parameters of the smoothed regions are determined, specifically including: Based on the body image data, smooth edge information of the body image data is identified. Using the smoothed edge information, we analyze the regions of change containing pixel gradient variations in the body image data. The changed regions are verified, and the verified changed regions are used as smoothed regions. The smoothing parameters of the smoothed regions are then extracted. Based on the smoothing parameter, the texture geometry features of wrinkles and depressions in the body image data are recovered using a deep neural network to obtain body reconstruction image data; After extracting the initial sparse feature points from the reconstructed body image data, a surface fitting model is used to fit the initial sparse feature points to obtain a three-dimensional human body model, specifically including: By monitoring the pixel displacement and deformation velocity in specific regions of the body in the reconstructed image data, pixel displacement information and deformation velocity information are obtained. Based on the pixel displacement information and deformation velocity information, local geometric features in the reconstructed body image data are obtained. Based on the local geometric features and preset topological constraints, deviation information is obtained. The degree of deviation between the body posture represented by the deviation information and the preset body shape topological constraints is evaluated. Based on the degree of deviation and the preset deviation threshold, after determining the deformation parameter boundary of the preset body shape topological constraint, the initial specific topological constraint is obtained. After performing a geometric continuity check on the initial specific topological constraints, the local discontinuities in the initial specific topological constraints are corrected to obtain the corrected specific topological constraints. Based on the specific topological constraints, after extracting the initial sparse feature points from the body reconstruction image data, the initial sparse feature points are fitted using a surface fitting model to obtain a three-dimensional human body model. The virtual underwear model is matched with the three-dimensional human body model to obtain model matching parameters; Using the model matching parameters, the virtual underwear model is modified to obtain the modified underwear parameters.

2. The method for virtual try-on design service for underwear according to claim 1, characterized in that, After performing geometric continuity verification on the initial specific topological constraints, correcting local discontinuities in the initial specific topological constraints, and obtaining the corrected specific topological constraints, the method further includes: Obtain the local curvature variation parameters of the modified specific topological constraint; Confirm the degree of matching between the corrected specific topological constraints and the specific body region represented by the deviation information; The weights of the energy function in the surface fitting model are determined based on local curvature variation parameters, matching degree, and preset matching threshold. Based on the weights of the energy function in the surface fitting model, the modified specific topological constraints are adjusted to obtain the adjusted specific topological constraints.

3. The method for virtual try-on design service for underwear according to claim 2, characterized in that, The step of adjusting the modified specific topological constraints based on the weights of the energy function in the surface fitting model to obtain the adjusted specific topological constraints includes: The body-specific regions represented by deviation information are divided, and the local feature stability information of each sub-region is determined. Based on the weights of the energy function in the surface fitting model, determine the adjustment step size of the weights; After evaluating the local feature stability information of each sub-region, if the evaluation of the local feature stability information is unqualified, the weight of the sub-region corresponding to the local feature stability information is adjusted by adjusting the step size to obtain the weight of each adjusted sub-region. Based on the weights of each adjusted sub-region, the modified specific topological constraints are adjusted to obtain the adjusted specific topological constraints.

4. The method for virtual try-on design service for underwear according to claim 3, characterized in that, The step of dividing the body-specific regions represented by deviation information and determining the local feature stability information of each sub-region includes: Based on the deviation information characterization of specific body regions, the texture density and curvature change gradient of specific body regions in the reconstructed body image data are determined; Based on the texture density and curvature change gradient, the granularity of sub-region division in specific body regions is adjusted. By utilizing the sub-regional granularity of the body-specific region, the body-specific region characterized by deviation information is divided, and the local feature stability information of each sub-region is determined.

5. The method for virtual try-on design service for underwear according to claim 1, characterized in that, The steps to obtain user-uploaded body image data and virtual underwear models include: Obtain initial body image data and virtual underwear model uploaded by the user; The initial body image data is denoised to obtain denoised initial body image data. After extracting the detail regions from the initial data of the denoised body image, local smoothing is performed on the detail regions, and global smoothing is performed on the non-detail regions in the initial data of the denoised body image to obtain the body image data.

6. A virtual fitting and design service system for underwear, used to implement the virtual fitting and design service method for underwear as described in claim 1, characterized in that, The system includes: The acquisition module is used to acquire body image data and virtual underwear models uploaded by users; The analysis module is used to analyze the smoothed regions in the body image data and then confirm the smoothing parameters of the smoothed regions. The recovery module is used to recover the texture geometry features of wrinkles and depressions in the body image data based on smoothing parameters and using a deep neural network to obtain body reconstruction image data; The fitting module is used to extract initial sparse feature points from the body reconstruction image data, and then use a surface fitting model to fit the initial sparse feature points to obtain a three-dimensional human body model. The matching module is used to match the virtual underwear model with the three-dimensional human body model to obtain model matching parameters; The design module is used to modify the virtual underwear model using the model matching parameters to obtain the modified underwear parameters.

Images