New chinese style clothing collocation system and method based on virtual fitting and style matching

Abstract

This application discloses a system and method for matching new Chinese-style clothing based on virtual try-on and style matching, belonging to the field of e-commerce. This application obtains basic user data, behavioral data, feedback data, and information on currently purchased items from an e-commerce platform. Based on the multi-dimensional characteristics of the new Chinese-style clothing items purchased by the user, it maps the clothing wearing scenario space, evaluates the user's corresponding wearing scenario, and selects a set of new Chinese-style clothing candidates that meet the user's scenario needs from the e-commerce platform's product library. It generates matching effects between purchased items and candidate items based on virtual try-on, evaluates the matching of purchased items and candidate items, assesses the user's focus on clothing attributes based on historical user behavior and evaluation feedback data, and comprehensively scores candidate items based on the user's wearing scenario, matching evaluation, and focus evaluation, and makes recommendations based on the comprehensive score, thereby improving matching coordination and the accuracy of user decision-making.

Description

Technical Field

[0001] This application falls under the field of e-commerce, specifically a new Chinese-style clothing matching system and method based on virtual try-on and style matching. Background Technology

[0002] With the continuous development of e-commerce technology, apparel has become an important part of online consumption. However, apparel has significant non-standardized characteristics; its wearing effect is influenced by various factors such as individual body shape, aesthetic preferences, wearing occasions, and matching relationships. Without the opportunity to try on clothes in person, consumers often find it difficult to accurately determine whether the clothing suits their needs, leading to uncertain purchasing decisions and high return rates. Existing apparel recommendation methods mostly focus on product recommendations based on collaborative filtering or simple content matching, primarily relying on users' historical purchase records, browsing behavior, or similar user preferences. They lack in-depth modeling of actual wearing scenarios and fail to reflect users' true wearing needs within a specific timeframe. For Neo-Chinese style clothing, its characteristics in style structure, cultural elements, craftsmanship details, and wearing occasions are significantly different from modern ready-to-wear, requiring even higher standards for matching wearing occasions and styles.

[0003] This application integrates user basic data, behavioral data, and feedback data, and introduces a clothing scene prediction mechanism based on clothing feature vectors to map new Chinese-style clothing to a quantifiable scene space. This allows the recommendation process to filter based on actual wearing occasions. At the same time, it combines virtual try-on technology to combine purchased items with candidate items, pays attention to user preferences, improves the matching degree between recommendation results and users' actual wearing needs, enhances user experience, and reduces the risk of returns and exchanges. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application proposes a new Chinese-style clothing matching system and method based on virtual try-on and style matching.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A new Chinese-style clothing matching system and method based on virtual try-on and style matching includes the following specific steps:

[0007] Based on the e-commerce platform, user basic data, behavioral data, feedback data, and information on currently purchased items are obtained;

[0008] Based on the multi-dimensional clothing characteristics of the new Chinese-style clothing items purchased by users, the clothing wearing scene space is mapped through a scene mapping model to evaluate the corresponding wearing scene of the user.

[0009] Based on the assessed wearing scenarios, a set of candidate new Chinese-style clothing that meets the user's scenario needs is selected from the product library of the business platform, forming a set of candidate items that match the user's current wearing scenario;

[0010] Based on virtual try-on, the system generates the matching effects of purchased items and candidate items, and evaluates the matching of purchased items and candidate items.

[0011] Based on user history and feedback data, we assess the user’s focus on clothing attributes.

[0012] Based on the user's wearing scenario, matching evaluation, and focus evaluation, candidate items are comprehensively scored, and recommendations are made based on the comprehensive score.

[0013] Preferably, the process of obtaining user basic data, behavioral data, feedback data, and information on currently purchased items based on the e-commerce platform includes the following specific steps:

[0014] S11. Obtain basic user data through e-commerce platforms, including age range, gender, height, weight, and 3D data. At the same time, obtain user behavior data, including historical purchase records, browsing history, favorites, shopping cart, search history, and page dwell time, to reflect the user's clothing habits and style preferences. Among them, the types of new Chinese-style clothing purchased in the past include Hanfu, modified cheongsam, and new Chinese-style everyday clothing. The browsing clothing attribute information includes style, color scheme, and fabric. The historical purchase records are all clothing transaction records completed by the user within a preset time window.

[0015] S12. Obtain user feedback data, including return and exchange records, size mismatch markings, and evaluation and feedback information. Extract keywords related to fabric, comfort, cultural elements, tailoring, and wearability from the evaluation text using natural language processing.

[0016] S13. Obtain the user's current purchase record for new Chinese-style clothing items that the user has completed payment for or entered the payment process for in this session or the most recent transaction. Store the item number, style category, price range, color, fabric, pattern, and product tag as structured data.

[0017] Preferably, the process of mapping the clothing wearing scenario space based on the multi-dimensional clothing features of the new Chinese-style clothing items purchased by the user, and evaluating the user's corresponding wearing scenario through a scenario mapping model, includes the following specific steps:

[0018] S21. Select the most recently purchased New Chinese style clothing item from the user's purchase records and extract the clothing feature vector corresponding to the New Chinese style clothing item. The clothing feature vector includes style and structure features, fabric and craftsmanship features, color and pattern features, and seasonal and climate features. The style and structure features include structural information such as stand-up collar, front opening, frog buttons, diagonal opening, modified cheongsam outline, and cuff design, which reflect the formality and traditional attributes of the clothing in the New Chinese style system. The fabric and craftsmanship features include fabric type, weaving process, embroidery and dyeing methods, which are used to judge the seasonal suitability, wearing comfort and etiquette level of the clothing. The color and pattern features include the main color, color saturation and content of traditional pattern elements, which are used to evaluate the clothing's tendency in cultural expression and visual style. The seasonal and climate features include the thickness of the clothing, the suitable season and regional climate conditions, which are used to avoid recommending clothing that is obviously incompatible with the current environment. After labeling all the features, a clothing feature vector is formed.

[0019] S22. Map the clothing feature vector to a predefined wearing scene space. The wearing scene space consists of multiple quantifiable scene feature dimensions. Use a large number of clothing images and text data with labeled scenes to train a multi-classification model. The input of the model is the features of the product, and the output is the probability distribution of the product belonging to each scene. Input the clothing feature vector into the multi-classification model to obtain the probability distribution of the scene to which the clothing belongs. Take the scene with the highest probability as the main prediction scene.

[0020] Preferably, the process of selecting a set of candidate new Chinese-style clothing that meets the user's needs based on the evaluated wearing scenario, and forming a set of candidate items that match the user's current wearing scenario, includes the following specific steps:

[0021] S31. Traverse all the new Chinese-style clothing items in the product library. Each item has a pre-labeled scene vector. The scene vector includes features such as formality, cultural element intensity, comfort, and applicable season. Calculate the scene similarity between the user scene vector and the scene vector of each item. The scene similarity calculation is achieved through cosine similarity.

[0022] S32. Filter individual items according to the set scene similarity threshold, and retain individual items that meet the similarity threshold to form a candidate individual item set.

[0023] Preferably, the step of generating the combination effect of purchased items and candidate items based on virtual try-on, and evaluating the combination of purchased items and candidate items, includes the following specific steps:

[0024] S41. Combine the purchased new Chinese-style clothing items with the candidate items on the user's virtual model to perform a virtual try-on and generate a matching image.

[0025] S42. Obtain the silhouette parameters of the most recently purchased Chinese-style clothing item and candidate items. These silhouette parameters include the vertical length ratio and horizontal looseness parameters of the clothing when worn. The vertical length ratio is obtained by comparing the garment length to the standard height, and the horizontal looseness is calculated by comparing the garment's circumference to the body's baseline circumference. Determine the combination relationship based on the positional attributes of the two garments in an outfit. The combination relationship can be a top-bottom combination or an outerwear-innerwear combination. When the combination relationship is a top-bottom combination, simultaneously calculate the length ratio between the top and bottom, comparing the length ratio value to a pre-set coordination range. If the ratio falls within the coordination range, set the ratio coordination score to the maximum baseline value; if the ratio deviates from the coordination range,... The proportional coordination score is calculated using a linear function based on the deviation of the garment from the interval boundary. Simultaneously, the difference in the looseness parameter between the two garments is calculated and compared with a preset looseness-tightness balance threshold range. When the difference in the looseness parameter is within the threshold range, it is determined to be a state of complementary looseness and tightness, and the looseness-tightness coordination score is set as the maximum baseline value. When the difference in the looseness parameter exceeds the threshold range, the score is reduced according to the extent of the excess. When the combination relationship is an outerwear-innerwear combination, the length difference and looseness difference between the outerwear and innerwear are calculated, and a layer coordination score is generated based on whether the preset coverage conditions are met. The proportional coordination score, looseness-tightness coordination score, and layer coordination score are normalized and weighted and summarized according to preset weights to obtain the silhouette coordination score.

[0026] S43. Obtain the color parameters of the most recently purchased new Chinese-style clothing item and candidate items. Map the main color, secondary color, and accent color of the two items to a unified color space, and extract the hue, lightness, and saturation values ​​respectively. Determine the color contrast relationship type based on the position of the clothing in the outfit combination, and calculate the hue difference, lightness difference, and saturation difference between the corresponding color features of the two items. Compare the hue difference, lightness difference, and saturation difference with a pre-defined color harmony range, wherein the hue difference is used to determine... Whether the color meets the rules of the same color family, adjacent colors, or contrasting colors, the brightness difference is used to judge whether the upper and lower layers are clear, and the saturation difference is used to judge whether the overall visual is too monotonous or too jarring. When the hue difference, brightness difference, and saturation difference all fall within the corresponding harmonious range, the color harmony score is set as the baseline maximum value. When any difference exceeds the corresponding range, the color harmony score is corrected according to the extent of the excess through a continuously decreasing function. The color harmony scores of the three dimensions of hue, brightness, and saturation are normalized and weighted to obtain the color matching score.

[0027] S44. The cultural elements of the most recently purchased Chinese-style clothing items and candidate items are structured and encoded. The cultural elements include collar type, opening method, button or knot structure, pattern type, embroidery or dyeing process, and traditional cultural symbol attributes. Each type of cultural element is represented by a discrete label. Based on the cultural attributes marked in the product information, cultural element feature vectors of candidate items and purchased items are constructed respectively. Each dimension of the vector is matched and calculated. When the cultural element matching value is within the preset threshold range, it is recorded as a matching state. When the cultural element matching value exceeds the preset threshold, it is recorded as a non-matching state. The matching ratio of the two garments in all cultural element dimensions is calculated. Combined with the importance weight of different cultural elements, the matching results are weighted and summarized to obtain a cultural element consistency score.

[0028] S45. The matching score of the candidate items is obtained by weighting and summing the silhouette coordination score, color matching score, and cultural element consistency score.

[0029] Preferably, the assessment of users' focus on clothing attributes based on historical user behavior and feedback data includes the following specific steps:

[0030] By acquiring users' historical purchase records, reviews, and feedback data, we can identify users' key concerns regarding clothing attributes. We obtain fabric attention weight by statistically analyzing the frequency of keywords related to fabric, comfort, and material in user reviews and their rating sensitivity. We obtain style attention weight by analyzing the proportion of different styles in historical purchases, repeat purchase preferences, and keywords in review texts. We obtain comfort attention weight by analyzing user feedback on fabric softness, fit, and discomfort in return and exchange records. We then form a weight vector from the fabric attention weight, style attention weight, and comfort attention weight, and use this weighted evaluation to assess the attribute vectors of candidate items to obtain a clothing attention score.

[0031] Preferably, the step of comprehensively scoring candidate items based on user wearing scenarios, matching evaluation, and focus evaluation, and making recommendations based on the comprehensive score, includes the following specific steps:

[0032] The candidate items are weighted and calculated based on scene similarity, matching score, and clothing focus score. The items are then sorted from highest to lowest score, and the K highest-scoring items are selected as the final recommendation results. The results are then displayed through virtual try-on.

[0033] A new Chinese-style clothing matching system based on virtual try-on and style matching is implemented based on the aforementioned new Chinese-style clothing matching method based on virtual try-on and style matching, specifically including:

[0034] The data acquisition module is used to obtain basic user data, behavioral data, feedback data, and information on currently purchased items from e-commerce platforms.

[0035] The dressing scenario assessment module is used to map the dressing scenario space based on the multi-dimensional clothing features of the new Chinese-style clothing items purchased by the user, and to assess the dressing scenario corresponding to the user.

[0036] The scenario similarity assessment module is used to filter out a set of new Chinese-style clothing candidates that meet the user's scenario needs from the product library of the business platform, forming a set of candidate items that match the user's current wearing scenario;

[0037] The matching evaluation module is used to generate matching effects of purchased items and candidate items through virtual try-on, and to evaluate the matching of purchased items and candidate items.

[0038] The focus assessment module is used to assess users' focus on clothing attributes by using their historical behavior and evaluation feedback data.

[0039] The comprehensive matching module is used to give a comprehensive score to candidate items based on the user's wearing scenario, matching evaluation, and focus evaluation, and then make clothing recommendations based on the comprehensive score.

[0040] An electronic device includes: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0041] The processor executes the aforementioned new Chinese-style clothing matching method based on virtual try-on and style matching by calling the computer program stored in the memory.

[0042] A computer-readable storage medium is characterized by storing instructions that, when executed on a computer, cause the computer to perform the aforementioned method for matching new Chinese-style clothing based on virtual try-on and style matching.

[0043] Compared with the prior art, the beneficial effects of this application are:

[0044] This application acquires basic user data, behavioral data, feedback data, and information on currently purchased items from e-commerce platforms. Based on the multi-dimensional characteristics of the purchased Neo-Chinese style clothing items, it maps the clothing wearing scenario space through a scenario mapping model, assesses the user's corresponding wearing scenario, and selects a set of Neo-Chinese style clothing candidates that meet the user's scenario needs from the e-commerce platform's product library. This forms a set of candidate items that match the user's current wearing scenario. Based on virtual try-on, it generates matching effects between purchased items and candidate items, evaluates the matching of purchased items and candidate items, assesses the user's focus on clothing attributes based on historical user behavior and evaluation feedback data, and comprehensively scores the candidate items based on the user's wearing scenario, matching evaluation, and focus evaluation. Recommendations are then made based on the comprehensive score, improving matching coordination and the accuracy of user decision-making. Attached Figure Description

[0045] Figure 1 This is a schematic diagram illustrating the overall process of the new Chinese clothing matching method based on virtual try-on and style matching in this application;

[0046] Figure 2 This is a schematic diagram of the scenario mapping for this application;

[0047] Figure 3 This is a flowchart of the comprehensive matching score calculation process for this application;

[0048] Figure 4 This is a schematic diagram of the overall framework of the new Chinese-style clothing matching system based on virtual try-on and style matching in this application. Detailed Implementation

[0049] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0050] Example 1

[0051] Please see Figures 1-3 One embodiment provided in this application is a new Chinese-style clothing matching method based on virtual try-on and style matching, which includes the following specific steps:

[0052] Based on the e-commerce platform, user basic data, behavioral data, feedback data, and information on currently purchased items are obtained;

[0053] Based on the multi-dimensional clothing characteristics of the new Chinese-style clothing items purchased by users, the clothing wearing scene space is mapped through a scene mapping model to evaluate the corresponding wearing scene of the user.

[0054] Based on the assessed wearing scenarios, a set of candidate new Chinese-style clothing that meets the user's scenario needs is selected from the product library of the business platform, forming a set of candidate items that match the user's current wearing scenario;

[0055] Based on virtual try-on, the system generates the matching effects of purchased items and candidate items, and evaluates the matching of purchased items and candidate items.

[0056] Based on user history and feedback data, we assess the user’s focus on clothing attributes.

[0057] Based on the user's wearing scenario, matching evaluation, and focus evaluation, candidate items are comprehensively scored, and recommendations are made based on the comprehensive score.

[0058] In this embodiment, it should be specifically explained that obtaining user basic data, behavioral data, feedback data, and information on currently purchased items based on the e-commerce platform includes the following specific steps:

[0059] S11. Obtain basic user data through e-commerce platforms, including age range, gender, height, weight, and 3D data. Simultaneously, obtain user behavior data, including historical purchase records, browsing history, favorites, shopping cart, search history, and page dwell time, to reflect the user's clothing habits and style preferences. Historical purchases of new Chinese-style clothing include Hanfu, modified cheongsam, and new Chinese-style everyday wear. Browsing clothing attribute information includes style (stand-up collar, frog buttons, double-breasted, etc.), color scheme (dark, solid, bright), and fabric (cotton, linen, silk, blends). Historical purchase records represent all purchases made by the user within a preset time window (the most recent year). Clothing transaction records include specific item information, purchase time, purchase frequency, repurchase status, and the proportion of different categories of new Chinese style clothing. Statistical analysis of historical purchase records is used to determine users' long-term style preferences and stable wearing habits. Browsing behavior records include users' clicks, dwell time, number of repeated visits, and actions such as adding to favorites or shopping carts but not making a purchase on new Chinese style clothing related product pages within the platform. Data is obtained through log collection, and different browsing records are assigned different weights based on browsing depth and behavior intensity to reflect users' potential interests rather than explicitly expressed purchase intentions.

[0060] S12. Obtain user feedback data, including return and exchange records, size mismatch markings, and evaluation and feedback information. Extract keywords related to fabric, comfort, cultural elements, tailoring, and wearability from the evaluation text through natural language processing to enhance the judgment of users' true preferences and reduce the bias caused by relying solely on purchase data.

[0061] For example, in this embodiment, natural language processing includes the following specific steps: when processing user evaluations and feedback information, the evaluation text is preprocessed, including word segmentation, removal of stop words, and normalization of synonyms, in order to eliminate the influence of differences in text expression on the analysis results. Based on a pre-constructed keyword dictionary for the clothing field, keyword matching and statistics are performed on the evaluation text. The keyword dictionary includes at least a set of words related to fabric attributes, wearing comfort, new Chinese cultural elements, tailoring structure, and practicality. In the keyword extraction process, the frequency of occurrence of keywords in a single evaluation and their cumulative occurrence in the user's historical evaluations are combined to assign corresponding weight values ​​to different keywords. When positive or negative sentiment modifiers appear simultaneously in the evaluation text, the weight of the corresponding keywords is enhanced or attenuated. The weighted statistical results of various keywords are summarized to form a user feedback feature vector, which is used to characterize the user's true attention to different clothing attribute dimensions.

[0062] S13. Obtain the user's current purchase record for new Chinese-style clothing items that the user has completed payment for or entered the payment process in this session or the most recent transaction. This reflects the user's clothing needs and aesthetic preferences in the current time period. Store the item number, style category, price range, color, fabric, pattern, and product tag as structured data.

[0063] In this embodiment, it should be specifically explained that, based on the multi-dimensional clothing features of the new Chinese-style clothing items purchased by the user, the clothing wearing scene space is mapped through a scene mapping model, and the evaluation of the user's corresponding wearing scene includes the following specific steps:

[0064] S21. Select the most recently purchased New Chinese-style clothing item from the user's purchase records. Compare it with historical purchase records. The most recently purchased item is closest to the user's current actual needs in terms of time, and its corresponding wearing purpose, occasion, and seasonal attributes have higher credibility. Analyze the clothing attributes of the most recently purchased item and extract the clothing feature vector corresponding to the New Chinese-style clothing item. The clothing feature vector includes style and structural features, fabric and craftsmanship features, color and pattern features, and seasonal and climate features. Among them, style and structural features include stand-up collar, front opening, frog buttons, diagonal placket, modified cheongsam silhouette, sleeves, etc. Structural information such as the design of the garment reflects the formality and traditional attributes of the clothing within the new Chinese style system. Fabric and craftsmanship characteristics include fabric type, weaving process, embroidery and dyeing methods, etc., used to judge the seasonal suitability, wearing comfort and etiquette level of the clothing. Color and pattern characteristics include main color, color saturation and content of traditional pattern elements, used to assess the tendency of the clothing in cultural expression and visual style. Seasonal and climate characteristics include the thickness of the clothing, the season it is suitable for and the regional climate conditions, used to avoid recommending clothing that is obviously incompatible with the current environment. After labeling all kinds of characteristics, a clothing feature vector is formed.

[0065] S22. Map the clothing feature vector to a predefined wearing scenario space. The wearing scenario space consists of multiple quantifiable scenario feature dimensions. The scenario feature dimensions are used to map clothing attributes to numerical usage occasion features, including formality, cultural display intensity, daily comfort, and social attribute intensity. Through the scenario mapping model, the scenario vector corresponding to the user is obtained. The mapping model is trained by the correspondence between historical clothing and actual wearing scenarios. The numerical values ​​of each dimension of the scenario vector are used to quantitatively describe the user's current wearing needs for new Chinese style clothing. A large amount of clothing images and text data with labeled scenarios are used to train a multi-classification model (e.g., a fusion model of text classification based on BERT and image classification based on ResNet). The input of the model is the features of the product, and the output is the probability distribution of the product belonging to each scenario. The clothing feature vector is input into the multi-classification model to obtain the probability distribution of the scene to which the clothing belongs. The scenario with the highest probability is taken as the predicted scenario.

[0066] For example, in this embodiment, the training of the multi-classification model includes the following specific steps: Extracting massive amounts of new Chinese-style clothing product data from an e-commerce platform database. Each data entry includes: main product image, title, detailed description, and attribute tags. Simultaneously, scene tags are defined and labeled. Fashion experts and data analysts jointly define a scene system suitable for new Chinese-style clothing. Preliminary scene labeling is performed using a keyword-based rule model. Labelers then correct and confirm the preliminary results based on the overall visual style of the product (style, color, pattern, material) and text description, forming a product-scene labeled dataset. A ResNet-50 model pre-trained on a large fashion dataset (DeepFashion) is used as the backbone. The network takes the main image of the product as input and extracts the global feature map of its last convolutional layer through ResNet-50. After passing through a global average pooling layer, a 512-dimensional or 2048-dimensional visual feature vector is obtained. A pre-trained Chinese BERT model is used to concatenate the product title and key description into a text segment, which is then input into the BERT model. The final hidden state marked with [CLS] is taken as the text feature vector. The visual feature vector and the text feature vector are fused through a cross-attention mechanism. When a user completes a purchase and obtains a single item, the online prediction process is initiated. The data is input into the fully connected layer and softmax layer of the pre-trained scene classifier, and the classifier outputs a scene probability distribution vector.

[0067] In this embodiment, it should be specifically explained that, based on the evaluated wearing scenario, the selection of a set of candidate new Chinese-style clothing that meets the user's scenario needs from the e-commerce platform's product library, forming a set of candidate items that match the user's current wearing scenario, includes the following specific steps:

[0068] S31. Traverse all the new Chinese-style clothing items in the product library. Each item has a pre-labeled scene vector. The scene vector includes features such as formality, cultural element intensity, comfort, and applicable season. Calculate the scene similarity between the user scene vector and the scene vector of each item. The scene similarity calculation is achieved through cosine similarity.

[0069] S32. Based on the set scene similarity threshold, filter individual items and retain those that meet or exceed the similarity threshold to form a candidate item set. The similarity threshold is an empirical threshold obtained by statistical analysis of historical recommendation click-through rates, purchase conversion rates, and user satisfaction. It can be weighted and adjusted according to different scene dimensions. By using scene similarity, we ensure that each piece of clothing in the candidate set is highly consistent with the user's current needs in the overall scene, thereby improving the accuracy of recommendations and user experience.

[0070] In this embodiment, it should be specifically explained that the evaluation of the combination of purchased items and candidate items based on virtual try-on includes the following specific steps:

[0071] S41. Combine the purchased new Chinese-style clothing items with the candidate items on the user's virtual model to perform a virtual try-on and generate a matching image.

[0072] S42. Silhouette Coordination measures whether the overall shape proportions of recommended items and purchased items are coordinated. For example, it considers the length ratio of tops and bottoms, and the pairing of loose and fitted silhouettes. It obtains the silhouette parameters of the most recently purchased Chinese-style clothing item and candidate items. These silhouette parameters include the vertical length ratio and horizontal looseness parameters of the clothing when worn. The vertical length ratio is obtained by comparing the garment length to a standard height, and the horizontal looseness is calculated by comparing the garment's circumference to a baseline body circumference. The combination relationship is determined based on the positional attributes of the two garments in the outfit. The combination relationship can be a top-bottom combination or an outerwear-innerwear combination. When the combination relationship is a top-bottom combination, the length ratio between the top and bottom is calculated simultaneously. The length ratio value is compared to a pre-set coordination range. When the ratio falls within the coordination range, the proportion coordination score is set to the maximum baseline value. When the proportion deviates from the coordination range, the proportion coordination score is calculated using a linear function based on the deviation from the range boundary. At the same time, the difference in the looseness parameter of the two garments is calculated and compared with a preset looseness-tightness balance threshold range. When the difference in the looseness parameter is within the threshold range, it is determined to be a state of complementary looseness and tightness, and the looseness-tightness coordination score is set as the maximum benchmark value. When the difference in the looseness parameter exceeds the threshold range, the score is reduced according to the extent of the excess. When the combination relationship is an outerwear-innerwear combination, the length difference and looseness difference between the outerwear and the innerwear are calculated, and a layer coordination score is generated based on whether the preset coverage conditions are met. The proportion coordination score, looseness-tightness coordination score, and layer coordination score are normalized and weighted and summarized according to preset weights to obtain the silhouette coordination score, which is used to characterize the silhouette coordination degree of the recommended single item and the purchased single item.

[0073] S43. Color Harmony Assessment: Recommend the pairing effect of the recommended single item with the purchased single item in terms of main color, secondary color, and contrast. Obtain the color parameters of the most recently purchased New Chinese Style clothing item and candidate items. Map the main color, secondary color, and accent color of the two garments to a unified color space. For example, convert RGB color values ​​to the HSV color space, which has stronger perceptual consistency, and extract hue, lightness, and saturation values ​​respectively. Determine the color contrast relationship type based on the position of the garments in the outfit combination, and calculate the hue difference, lightness difference, and saturation difference between the corresponding color characteristics of the two garments. Compare these hue difference, lightness difference, and saturation difference with a pre-defined color harmony range. In the comparison, the hue difference is used to determine whether the color meets the rules of the same color family, adjacent colors, or contrasting colors; the lightness difference is used to determine whether the upper and lower layers are clear; and the saturation difference is used to determine whether the overall visual effect is too monotonous or too jarring. When the hue difference, lightness difference, and saturation difference all fall within the corresponding harmony range, the color harmony score is set as the baseline maximum value. When any difference exceeds the corresponding range, the color harmony score is corrected according to the extent of the excess through a continuously decreasing function. The color harmony scores of the three dimensions of hue, lightness, and saturation are normalized and weighted to obtain the color matching score, which is used to characterize the color harmony evaluation result of the recommended item and the purchased item in the overall outfit.

[0074] S44. Cultural Element Consistency Assessment: Recommended items are assessed for their compatibility with purchased items in terms of traditional patterns, embroidery, frog buttons, collar styles, and other Neo-Chinese elements to ensure overall style consistency. The Neo-Chinese cultural elements contained in the most recently purchased Neo-Chinese clothing item and candidate items are structurally coded. These cultural elements include collar style, opening method, frog button or knot structure, pattern type, embroidery or dyeing techniques, and traditional cultural symbol attributes. Each type of cultural element is represented by a discrete tag. Based on the cultural attributes marked in the product information or the cultural features extracted through image and text analysis, candidate items are constructed accordingly. The cultural element feature vectors of selected items and purchased items are compared, and each dimension of the vector is matched and calculated. When the cultural element matching value is within a preset threshold range, it is recorded as a matching state. When the cultural element matching value exceeds the preset threshold, it is recorded as a non-matching state. The matching ratio of the two garments in all cultural element dimensions is calculated. The matching results are weighted and summarized by combining the importance weights of different cultural elements to obtain a cultural element consistency score. The importance weight can be determined based on the historical acceptance of wearing the garment. After normalization, the score is used to quantitatively characterize the degree of unity between the recommended items and purchased items in terms of the expression of New Chinese culture.

[0075] S45. The silhouette coordination score, color matching score, and cultural element consistency score are weighted and summed to obtain the matching score of the candidate items. Through matching evaluation, the situation where the recommended items conflict with the user's existing clothing style is avoided, thereby improving user acceptance and the coordination of actual wearing.

[0076] In this embodiment, it should be specifically explained that the evaluation of users' focus on clothing attributes based on users' historical behavior and evaluation feedback data includes the following specific steps:

[0077] By acquiring users' historical purchase records, reviews, and feedback data, we can identify users' key concerns regarding clothing attributes. We obtain fabric attention weight by analyzing the frequency and rating sensitivity of keywords related to fabric, comfort, and material in user reviews, reflecting the degree of importance users place on clothing fabric. We obtain style attention weight by analyzing the proportion of different styles in historical purchases, repeat purchase preferences, and keywords in review text, reflecting the degree of importance users place on style and cut. We obtain comfort attention weight by analyzing user feedback on fabric softness, fit, and discomfort in return / exchange records, representing users' concern for clothing comfort. We then combine the weights of fabric attention, style attention, and comfort attention... Weights are used to form a weight vector, which is then used to evaluate the attribute vectors of candidate items (numerical feature vectors of candidate items in dimensions such as fabric, style, cultural elements, and comfort) to obtain an apparel attention score. The higher the apparel attention score, the more the candidate item matches the user's personalized attention. By comprehensively analyzing the user's historical purchase records, reviews, and feedback data, the recommendation process reflects the personalized preferences that users have formed over a long period of time, avoiding the bias caused by recommending based solely on the current scenario or a single matching rule. At the same time, it distinguishes the differences in the degree of attention that different users pay to apparel attributes, and further prioritizes recommending candidate items that perform well in the attributes that users focus on, thereby improving the matching degree between the recommendation results and the user's actual needs.

[0078] In this embodiment, it is necessary to specifically explain that the process of comprehensively scoring candidate items based on user wearing scenarios, matching evaluations, and focus evaluations, and then making recommendations based on these comprehensive scores, includes the following specific steps:

[0079] The candidate items are weighted and calculated to obtain a comprehensive matching score based on the scene similarity, matching score, and clothing focus score. The higher the comprehensive matching score, the more the item matches the user's current scene, matching coordination, and personalized attention needs. After the comprehensive score is calculated, the items are sorted from high to low, and the K items with the highest scores are selected as the final recommendation results. The effect is then displayed through virtual try-on.

[0080] The advantages of this embodiment compared to the prior art are:

[0081] This application acquires basic user data, behavioral data, feedback data, and information on currently purchased items from e-commerce platforms. Based on the multi-dimensional characteristics of the purchased Neo-Chinese style clothing items, it maps the clothing wearing scenario space through a scenario mapping model, assesses the user's corresponding wearing scenario, and selects a set of Neo-Chinese style clothing candidates that meet the user's scenario needs from the e-commerce platform's product library. This forms a set of candidate items that match the user's current wearing scenario. Based on virtual try-on, it generates matching effects between purchased items and candidate items, evaluates the matching of purchased items and candidate items, assesses the user's focus on clothing attributes based on historical user behavior and evaluation feedback data, and comprehensively scores the candidate items based on the user's wearing scenario, matching evaluation, and focus evaluation. Recommendations are then made based on the comprehensive score, improving matching coordination and the accuracy of user decision-making.

[0082] Example 2

[0083] like Figure 4 As shown, the new Chinese-style clothing matching system based on virtual try-on and style matching is implemented based on the aforementioned method for matching new Chinese-style clothing based on virtual try-on and style matching. Specifically, it includes a data acquisition module, a wearing scenario evaluation module, a scenario similarity evaluation module, a matching evaluation module, a focus evaluation module, and a comprehensive matching module. The data acquisition module is used to obtain basic user data, behavioral data, feedback data, and information on currently purchased items from the e-commerce platform. The wearing scenario evaluation module is used to map the wearing scenario space using the multi-dimensional clothing features of the new Chinese-style clothing items purchased by the user, and evaluate the user's corresponding wearing scenario. The similarity assessment module is used to filter out a set of candidate new Chinese-style clothing that meets the user's needs in the business platform's product library, forming a set of candidate items that match the user's current wearing scenario; the matching assessment module is used to generate matching effects between purchased items and candidate items through virtual try-on, and to evaluate the matching of purchased items and candidate items; the focus assessment module is used to evaluate the user's focus on clothing attributes based on the user's historical behavior and evaluation feedback data; the comprehensive matching module is used to give a comprehensive score to candidate items based on the user's wearing scenario, matching assessment, and focus assessment, and to recommend clothing based on the comprehensive score.

[0084] Example 3

[0085] This embodiment provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0086] The processor executes the aforementioned method of matching new Chinese-style clothing based on virtual try-on and style matching by calling computer programs stored in memory.

[0087] The electronic device can vary considerably depending on its configuration or performance. It may include one or more Central Processing Units (CPUs) and one or more memories, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the novel Chinese clothing matching method based on virtual try-on and style matching provided in the above-described embodiment. The electronic device may also include other components for implementing its functions; for example, it may have wired or wireless network interfaces and input / output interfaces for data input and output. Details will not be elaborated upon in this embodiment.

[0088] Example 4

[0089] This embodiment proposes a computer-readable storage medium on which an erasable and rewritable computer program is stored.

[0090] When the computer program runs on the computer device, it causes the computer device to execute the aforementioned method of matching new Chinese clothing based on virtual try-on and style matching.

[0091] For example, computer-readable storage media can be read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage devices.

[0092] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired network and / or wireless network. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

Claims

1. A new method for matching Chinese-style clothing based on virtual try-on and style matching, characterized in that, It includes the following specific steps: Based on the e-commerce platform, user basic data, behavioral data, feedback data, and information on currently purchased items are obtained; Based on the multi-dimensional clothing characteristics of the new Chinese-style clothing items purchased by users, the clothing wearing scene space is mapped through a scene mapping model to evaluate the corresponding wearing scene of the user. Based on the assessed wearing scenarios, a set of candidate new Chinese-style clothing that meets the user's scenario needs is selected from the product library of the business platform, forming a set of candidate items that match the user's current wearing scenario; Based on virtual try-on, the system generates the matching effects of purchased items and candidate items, and evaluates the matching of purchased items and candidate items. Based on user history and feedback data, we assess the user’s focus on clothing attributes. Based on the user's wearing scenario, matching evaluation, and focus evaluation, candidate items are comprehensively scored and recommended. The specific steps include: calculating a comprehensive matching score by weighting the scenario similarity, matching score, and clothing focus score of the candidate items. The higher the score, the more the item matches the user's current scenario, matching coordination, and personalized focus needs. After the comprehensive score is calculated, the items are sorted from high to low, and the K items with the highest scores are selected as the final recommendation results. The effect is then displayed through virtual try-on.

2. The method for matching new Chinese-style clothing based on virtual try-on and style matching as described in claim 1, characterized in that, The process of obtaining user basic data, behavioral data, feedback data, and information on currently purchased items based on e-commerce platforms includes the following specific steps: S11. Obtain basic user data through e-commerce platforms, including age range, gender, height, weight, and 3D data. At the same time, obtain user behavior data, including historical purchase records, browsing history, favorites, shopping cart, search history, and page dwell time. S12. Obtain user feedback data, including return and exchange records, size mismatch markings, and evaluation and feedback information. Extract keywords related to fabric, comfort, cultural elements, tailoring, and wearability from the evaluation text using natural language processing. S13. Obtain the user's current purchase record, including information on the new Chinese-style clothing items that the user has paid for in their most recent transaction.

3. The method for matching new Chinese-style clothing based on virtual try-on and style matching as described in claim 2, characterized in that, The multi-dimensional clothing features based on the new Chinese-style clothing items purchased by the user, through a scene mapping model to map the clothing wearing scene space, and to evaluate the user's corresponding wearing scene, include the following specific steps: S21. Select the most recently purchased Chinese-style clothing item from the user's purchase records and extract the clothing feature vector corresponding to the item. The clothing feature vector includes style and structure features, fabric and craftsmanship features, color and pattern features, and seasonal and climate features. Style and structure features include stand-up collar, front opening, frog buttons, diagonal opening, modified cheongsam outline, and cuff design structure information. Fabric and craftsmanship features include fabric type, weaving process, embroidery and dyeing methods. Color and pattern features include main color tone, color saturation, and content of traditional pattern elements. Seasonal and climate features include clothing thickness, suitable season, and regional climate conditions. After labeling all features, a clothing feature vector is formed. S22. Map the clothing feature vector to a predefined wearing scene space. The wearing scene space consists of multiple quantifiable scene feature dimensions. Use a large number of clothing images and text data with labeled scenes to train a multi-classification model. The input of the model is the features of the product, and the output is the probability distribution of the product belonging to each scene. Input the clothing feature vector into the classification model to obtain the probability distribution of the scene to which the clothing belongs, and take the scene with the highest probability as the predicted scene.

4. The method for matching new Chinese-style clothing based on virtual try-on and style matching as described in claim 3, characterized in that, The process of selecting a set of candidate new Chinese-style clothing that meets the user's needs based on the assessed wearing scenario, and forming a set of candidate items that match the user's current wearing scenario, includes the following specific steps: S31. Traverse all the new Chinese-style clothing items in the product library. Each item has a pre-labeled scene vector. Calculate the scene similarity between the user scene vector and the scene vector of each item. The scene similarity calculation is achieved through cosine similarity. S32. Filter individual items according to the set scene similarity threshold, and retain individual items that meet the similarity threshold to form a candidate individual item set.

5. The new Chinese-style clothing matching method based on virtual try-on and style matching as described in claim 4, characterized in that, The process of generating styling effects for purchased and candidate items based on virtual try-on, and evaluating the styling of purchased and candidate items, includes the following specific steps: S41. Combine the purchased new Chinese-style clothing items with the candidate items on the user's virtual model to perform a virtual try-on and generate a matching image. S42. Obtain the silhouette parameters of the most recently purchased Chinese-style clothing item and candidate items. These silhouette parameters include the vertical length ratio and horizontal looseness parameters of the clothing when worn. Determine the combination relationship based on the positional attributes of the two garments in an outfit. The combination relationship can be a top-bottom combination or an outerwear-innerwear combination. When the combination relationship is a top-bottom combination, simultaneously calculate the length ratio between the top and bottom, comparing the length ratio value to a pre-defined coordination range. If the ratio falls within the coordination range, set the ratio coordination score to the maximum baseline value. If the ratio deviates from the coordination range, calculate the ratio coordination using a linear function based on the deviation from the range boundary. The scoring process involves calculating the difference in the looseness parameters of two garments and comparing it with a preset looseness-tightness balance threshold range. When the difference in the looseness parameters is within the threshold range, it is determined to be a state of complementary looseness and tightness, and the looseness-tightness coordination score is set to the maximum baseline value. When the difference in the looseness parameters exceeds the threshold range, the score is reduced according to the extent of the excess. When the combination relationship is an outerwear-innerwear combination, the length difference and looseness difference between the outerwear and the innerwear are calculated, and a layer coordination score is generated based on whether the preset coverage conditions are met. The proportion coordination score, looseness-tightness coordination score, and layer coordination score are normalized and weighted according to preset weights to obtain the silhouette coordination score. S43. Obtain the color parameters of the most recently purchased new Chinese-style clothing item and candidate items. Map the main color, secondary color, and partial accent color of the two garments to a unified color space, and extract the hue, lightness, and saturation values ​​respectively. Determine the color contrast relationship type based on the position of the garments in the outfit combination, and calculate the hue difference, lightness difference, and saturation difference between the corresponding color features of the two garments. Compare the hue difference, lightness difference, and saturation difference with the pre-set color harmony intervals. When the hue difference, lightness difference, and saturation difference all fall within the corresponding harmony interval, the color harmony score is set as the baseline maximum value. When any difference exceeds the corresponding interval, the color harmony score is corrected according to the excess range using a continuously decreasing function. The color harmony scores of the three dimensions of hue, lightness, and saturation are normalized and weighted to obtain the color matching score. S44. Structure the New Chinese cultural elements contained in the most recently purchased New Chinese style clothing items and candidate items. Each cultural element is represented by a discrete label. Based on the cultural attributes marked in the product information, construct cultural element feature vectors for both candidate and purchased items. Perform item-by-item matching calculations on each dimension of the vectors. When the cultural element matching value is within a preset threshold range, it is considered a matching state; when the cultural element matching value exceeds the preset threshold, it is considered a non-matching state. Calculate the matching ratio of the two garments across all cultural element dimensions. Combine the importance weights of different cultural elements to perform a weighted summary of the matching results and obtain a cultural element consistency score. S45. The matching score of the candidate items is obtained by weighting and summing the silhouette coordination score, color matching score and cultural element consistency score.

6. The new Chinese-style clothing matching method based on virtual try-on and style matching as described in claim 5, characterized in that, The assessment of users' focus on clothing attributes based on historical user behavior and feedback data includes the following specific steps: By acquiring users' historical purchase records, reviews, and feedback data, we can identify users' key concerns regarding clothing attributes. We obtain fabric attention weight by statistically analyzing the frequency of keywords related to fabric, comfort, and material in user reviews and their rating sensitivity. We obtain style attention weight by analyzing the proportion of different styles in historical purchases, repeat purchase preferences, and keywords in review texts. We obtain comfort attention weight by analyzing user feedback on fabric softness, fit, and discomfort in return and exchange records. We then form a weight vector from the fabric attention weight, style attention weight, and comfort attention weight, and use this weighted evaluation to assess the attribute vectors of candidate items to obtain a clothing attention score.

7. A new Chinese-style clothing matching system based on virtual try-on and style matching, implemented based on the new Chinese-style clothing matching method based on virtual try-on and style matching as described in any one of claims 1-6, characterized in that, Specifically, it includes: The data acquisition module is used to obtain basic user data, behavioral data, feedback data, and information on currently purchased items from e-commerce platforms. The dressing scenario assessment module is used to map the dressing scenario space based on the multi-dimensional clothing features of the new Chinese-style clothing items purchased by the user, and to assess the dressing scenario corresponding to the user. The scenario similarity assessment module is used to filter out a set of new Chinese-style clothing candidates that meet the user's scenario needs from the product library of the business platform, forming a set of candidate items that match the user's current wearing scenario; The matching evaluation module is used to generate matching effects of purchased items and candidate items through virtual try-on, and to evaluate the matching of purchased items and candidate items. The focus assessment module is used to assess users' focus on clothing attributes by using their historical behavior and evaluation feedback data. The comprehensive matching module is used to give a comprehensive score to candidate items based on the user's wearing scenario, matching evaluation, and focus evaluation, and then make clothing recommendations based on the comprehensive score.

8. An electronic device, comprising: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor is characterized in that it executes the new Chinese clothing matching method based on virtual try-on and style matching as described in any one of claims 1-6 by calling a computer program stored in the memory.

9. A computer-readable storage medium, characterized in that, The device stores instructions that, when executed on a computer, cause the computer to perform the new Chinese clothing matching method based on virtual try-on and style matching as described in any one of claims 1-6.

Images