Beautifying soft sweet formula recommendation method and system based on facial skin state analysis

Abstract

The invention belongs to the technical field of beauty soft sweets, and provides a beauty soft sweet formula recommendation method and system based on facial skin state analysis. The method comprises the following steps: acquiring multi-angle two-dimensional facial skin images of a user by using a mobile phone camera, and synthesizing the two-dimensional facial skin images into a three-dimensional facial skin image; identifying facial fine line data and facial color spot data from the three-dimensional facial skin image by using a feature extraction model based on infinity-former to form facial skin state data; inputting the facial skin state data into a rating analysis model, connecting a facial skin database with the rating analysis model, and outputting a facial fine line grade and a facial color spot grade; and matching with a formula database to obtain a target formula which contains the dosages of collagen and antioxidant in the beautifying soft sweets. According to the invention, the personalized beauty soft sweet recommendation formula can be conveniently generated.

Description

Technical Field

[0001] The present invention relates to the technical field of cosmetic soft candies, and in particular to a method and system for recommending cosmetic soft candy formulas based on facial skin condition analysis. Background Art

[0002] As demand for beauty products continues to grow, facial skin condition has become a crucial indicator of personal health and image. Skin problems like wrinkles and dark spots not only affect appearance but also reflect the extent of skin aging and damage, prompting consumers to continuously seek effective solutions. Gummy candies, due to their portability and fun appeal, have become an emerging hotspot in the beauty and wellness market. They contain collagen that enhances skin elasticity and antioxidants that combat free radical damage, demonstrating significant potential in the skin care sector.

[0003] However, the traditional beauty gummy market is plagued by a one-size-fits-all product supply model. Existing products often utilize fixed formulas, making it difficult to precisely tailor their formulations to the individual skin needs of individual consumers. Due to significant differences in individual skin types, age, living environments, and other factors, the collagen and antioxidant content in a single beauty gummy cannot meet the individual needs of all consumers. For example, those with more fine lines may require a higher dose of collagen to promote cell repair, while those with noticeable pigmentation may require more antioxidant supplements. Traditional formulas struggle to achieve this precise adjustment.

[0004] At the same time, traditional methods for skin condition monitoring and product recommendations have numerous limitations. Consumers have traditionally relied on professional skin testing equipment at beauty salons, but the high cost and time constraints limit testing frequency. Recommendations based on questionnaires or self-descriptions lack objective data support, resulting in subjectivity and low accuracy. These methods are unable to quantify the severity of skin issues like fine lines and dark spots, making it difficult to scientifically determine the appropriate dosage of various active ingredients in beauty gummies.

[0005] Therefore, how to improve the convenience and efficiency of providing users with personalized beauty gummy recommendation formulas is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] In this regard, the present invention provides a method, system, electronic device, computer storage medium and computer program product for recommending a beauty soft candy formula based on facial skin condition analysis to solve at least one of the above technical problems.

[0007] In a first aspect, the present invention provides a method for recommending a cosmetic soft candy formula based on facial skin condition analysis, comprising the following steps: The mobile phone camera is used to collect two-dimensional facial skin images of the user from multiple angles, and the two-dimensional facial skin images are synthesized into a three-dimensional facial skin image.

[0008] A feature extraction model based on ∞-former is used to identify facial fine line data and facial color spot data from the three-dimensional facial skin image to form facial skin status data; the facial skin status data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model, and the rating analysis model outputs the facial fine line level and the facial color spot level.

[0009] Based on the facial fine line level and the facial spot level, the recommended target formula is obtained by matching with a formula database. The target formula includes the dosage of collagen and antioxidants in the beauty gummy.

[0010] In a second aspect, the present invention provides a cosmetic soft candy recipe recommendation system based on facial skin condition analysis. The system includes a controller and a storage medium. The storage medium stores a computer program. The controller calls and executes the computer program to implement: The mobile phone camera is used to collect two-dimensional facial skin images of the user from multiple angles, and the two-dimensional facial skin images are synthesized into a three-dimensional facial skin image.

[0011] A feature extraction model based on ∞-former is used to identify facial fine line data and facial color spot data from the three-dimensional facial skin image to form facial skin status data; the facial skin status data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model, and the rating analysis model outputs the facial fine line level and the facial color spot level.

[0012] Based on the facial fine line level and the facial spot level, the recommended target formula is obtained by matching with a formula database. The target formula includes the dosage of collagen and antioxidants in the beauty gummy.

[0013] In a third aspect of the present invention, an electronic device is provided, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements any of the methods described above when executed by the processor.

[0014] According to a fourth aspect of the present invention, a computer storage medium is provided, wherein the computer storage medium stores a computer program executable by a processor to implement any of the methods described above.

[0015] According to a fifth aspect of the present invention, a computer program product is provided, which comprises a computer program executable by a processor to implement any of the methods described above.

[0016] This method uses a mobile phone camera to capture multi-angle images to create a three-dimensional facial skin image. It then uses an ∞-former feature extraction model to accurately identify fine lines and spots. Through rating analysis and formula matching, a personalized beauty gummy formula is generated. This method enables convenient and efficient skin condition detection and customized nutritional supplementation, breaking through the limitations of the traditional "one-size-fits-all" model and enhancing the targeted and effective skincare effects of beauty gummy candies to meet the individual needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of a method for recommending a beauty soft candy formula based on facial skin condition analysis disclosed in an embodiment of the present invention; Figure 2 It is a structural diagram of the feature extraction model disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following specific embodiments illustrate the implementation of this application. Those familiar with the art can easily understand the other advantages and functions of this application from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0021] like Figure 1 As shown, the embodiment of the present invention discloses a method for recommending a cosmetic soft candy formula based on facial skin condition analysis, comprising the following steps: S10, using a mobile phone camera to collect two-dimensional facial skin images of the user from multiple angles, and synthesizing the two-dimensional facial skin images into a three-dimensional facial skin image.

[0022] Mobile phone cameras are commonly used image acquisition devices, offering flexibility and convenience. By controlling the phone camera to capture the user's facial skin from different angles (such as the front, left and right sides), multiple 2D facial skin images are acquired. These multi-angle images are then fused using computer graphics and image processing techniques, such as algorithms based on feature point matching and stereo vision principles. Corresponding points in the different images (such as those at the corners of the eyes and wings of the nose) are matched and calibrated, establishing coordinate relationships in 3D space. This allows for the synthesis of a 3D facial skin image that faithfully reflects the facial skin's surface morphology and texture.

[0023] Compared with two-dimensional images from a single angle, three-dimensional facial skin images can present the three-dimensional structure and detailed information of the skin more comprehensively and accurately. They can capture subtle changes in facial skin at different angles, such as skin folds and uneven areas, avoiding information loss or misjudgment due to the limited viewing angle of two-dimensional images.

[0024] It is understood that the solution of the present invention is mainly used in medical or beauty institutions. The staff of the medical or beauty institution uses a mobile phone camera to capture the user's facial image, derives a personalized beauty soft candy formula according to the solution of the present invention, and then transmits the beauty soft candy formula to the manufacturer via a special APP, and the manufacturer produces the customized beauty soft candy. Of course, the solution of the present invention can also be used by the user to take a selfie using the mobile phone camera. Similarly, the beauty soft candy formula is transmitted to the manufacturer via a special APP, and the manufacturer produces the customized beauty soft candy. The present invention is not specifically limited to this.

[0025] S20, using an ∞-former-based feature extraction model to identify facial fine line data and facial color spot data from the three-dimensional facial skin image to form facial skin status data; inputting the facial skin status data into a rating analysis model, and connecting the facial skin database with the rating analysis model, the rating analysis model outputs the facial fine line level and the facial color spot level.

[0026] This paper constructs a feature extraction model based on the ∞-former. This feature extraction model performs in-depth analysis of three-dimensional facial skin images, leveraging the ∞-former's unique continuous spatial attention mechanism and viscous memory properties to accurately capture skin details. By continuously tracking image features such as texture, color, and shape, the feature extraction model can identify the number, length, depth, and direction of facial fine lines, which constitute facial fine line data; as well as the size, shape, color, and distribution of facial spots, which constitute facial spot data. After vectorization, the facial fine line and facial spot data constitute facial skin condition data. Reference Figure 2As shown, the feature extraction model of the present invention is mainly composed of the following key structures: Input layer: This layer receives 3D facial skin image data. These images undergo preprocessing, such as normalization and scaling, to ensure they have a uniform format and range for subsequent model processing.

[0027] Continuous Spatial Attention Mechanism: This is one of the core modules of the ∞-former and is used to represent input image information as a continuous signal. For example, it uses a linear combination of radial basis functions to determine the attention weights for each region in the image, highlighting key feature areas such as fine lines and spots. This module can process and store long sequences of information with a fixed computational load. Through the continuous spatial attention mechanism, it can more effectively capture the connections between different regions in the image, and is particularly capable of capturing detailed features such as fine lines and spots on the face.

[0028] Viscous Memory Module: This module is used to endow the feature extraction model with long-term memory capabilities. Based on the calculation results of the Continuous Spatial Attention Mechanism, this module updates and strengthens the memory of key features. New feature information is promptly incorporated into the memory, and already memorized features are dynamically adjusted based on their importance.

[0029] Feature extraction layer: A multi-layer neural network extracts features from the data processed by the continuous spatial attention module and the viscous memory module. Each layer learns different levels of features, from low-level edge and color features to high-level texture and shape features. Ultimately, these features are converted into vector representations suitable for subsequent analysis.

[0030] Output layer: The output layer outputs the extracted facial fine line data and facial spot data. The facial fine line data includes the number, length, depth, and direction of fine lines, and the facial spot data includes detailed information such as the size, shape, color, and distribution area of the spots.

[0031] Facial skin condition data is input into the rating analysis model, which then connects to a facial skin database for data comparison and evaluation. The database contains a large amount of labeled data on different skin conditions and their corresponding grading standards (i.e., labels). Based on these grading standards, the rating analysis model calculates the similarity or difference between the input data and the characteristics of each grade in the database to grade the user's facial fine lines and spots, ultimately outputting a quantified grade for each.

[0032] S30, matching the facial fine line level and the facial spot level with a formula database to obtain a recommended target formula, wherein the target formula includes the dosage of collagen and antioxidants in the beauty gummy.

[0033] The formula database has pre-integrated a large amount of research data and practical experience, and established a correspondence between different levels of facial fine lines and facial pigmentation and the formula of beauty gummies. That is, it records in detail the optimal dosage ratio of collagen and antioxidants in beauty gummies for various combinations of fine line levels and pigmentation levels.

[0034] The rating analysis model outputs the levels of facial fine lines and facial pigmentation as search criteria, and a precise match is performed in the formula database. Based on the input level information, the database screens out the corresponding recommended formula for the best beauty gummy candy and determines the specific dosage of collagen and antioxidants in the target formula. For example, for users with high fine line levels and low pigmentation levels, the formula database will match a formula with a high collagen content and a moderate antioxidant content, thereby customizing the beauty gummy candy formula that best suits their skin condition and achieving a personalized skin care plan. Of course, the recommended target formula still needs to be reviewed by relevant personnel before product production can be carried out to avoid the use of undesirable formulas.

[0035] This method uses a mobile phone camera to capture multi-angle images to create a three-dimensional facial skin image. It then uses an ∞-former feature extraction model to accurately identify fine lines and spots. Through rating analysis and formula matching, a personalized beauty gummy formula is generated. This method enables convenient and efficient skin condition detection and customized nutritional supplementation, breaking through the limitations of the traditional "one-size-fits-all" model and enhancing the targeted and effective skincare effects of beauty gummy candies to meet the individual needs of users.

[0036] As an example, the method of collecting multi-angle two-dimensional facial skin images of a user using a mobile phone camera includes: Using a mobile phone camera to capture a two-dimensional facial skin image of the user's front face, a rough rating of the user's facial skin condition is performed based on the coarse features of facial fine lines and facial spots in the two-dimensional facial skin image, and a scanning and shooting frequency is determined based on the rough rating results; A formal scanning prompt signal is output, and two-dimensional facial skin images of the user at multiple angles are collected according to the scanning shooting frequency.

[0037] When synthesizing a 3D facial skin image using multiple 2D facial skin images taken at different angles, theoretically, the more 2D facial skin images used for synthesis, the more realistic skin details the resulting 3D facial skin image will contain. This helps to improve the accuracy of the extracted facial fine lines and facial spots data relative to the actual skin condition. However, excessively high scanning and shooting frequencies can lead to significant increases in the capture and image processing load, and significantly prolong the time required to generate the 3D facial skin image. Therefore, it is necessary to determine an appropriate scanning and shooting frequency.

[0038] Specifically, the system first uses a mobile phone camera to capture a two-dimensional frontal facial skin image of the user, capturing image data containing preliminary information about fine lines and pigmentation. From this frontal facial image, it then extracts distinct, relatively macroscopic features of fine lines and pigmentation (i.e., coarse features), such as the approximate number of fine lines and the approximate area of pigmentation. Based on these coarse features, it then assigns a preliminary, broad rating to the user's facial skin condition, which can be categorized into several levels, such as mild, moderate, and severe, to quickly assess the severity of the user's skin issues.

[0039] The rough rating then determines the frequency of subsequent multi-angle image acquisition. If the rough rating is mild, the scan frequency is set relatively low, meaning that fewer angles of image acquisition will be sufficient for subsequent analysis. If the rough rating is moderate or severe, a higher scan frequency is required to capture 2D facial skin images from more angles to obtain more comprehensive skin information.

[0040] After determining the appropriate scanning frequency, the phone prompts the user or operator to begin the formal multi-angle image acquisition process. The user or operator then slowly moves the phone's camera around the user's face. During this slow movement, the phone captures 2D facial skin images from different angles (such as the front, left and right sides) at the previously determined scanning frequency.

[0041] This setting can ensure that the collected two-dimensional facial skin images are sufficient, so that the subsequently synthesized three-dimensional facial skin images contain enough skin details, while also avoiding unnecessary computational load due to excessive collection of two-dimensional facial skin images.

[0042] As an example, the feature extraction model includes an input layer, a continuous spatial attention mechanism module, a viscous memory module, a feature extraction layer, and an output layer; using the ∞-former-based feature extraction model to identify facial fine line data and facial spot data from the three-dimensional facial skin image includes: Segmenting the three-dimensional facial skin image into a plurality of local image blocks, and inputting each of the local image blocks into the input layer; The continuous spatial attention mechanism module simultaneously performs continuous signal conversion and attention weight calculation on multiple local image blocks through parallel computing; and when the amount of feature data output by the continuous spatial attention mechanism module exceeds the dynamic feature screening threshold, the viscous memory module preferentially retains feature data strongly related to facial fine lines and spots to obtain an intermediate feature data matrix; The feature extraction layer uses a lightweight neural network structure to extract and separate facial fine line sub-data and facial color spot sub-data corresponding to each local image block from the intermediate feature data matrix; The output layer summarizes each of the facial fine line sub-data and each of the facial color spot sub-data, and integrates them to generate the facial fine line data and the facial color spot data.

[0043] To more efficiently process 3D facial skin images, they are first segmented into multiple smaller local image patches to reduce computational complexity. Each segmented local image patch is then fed sequentially into the input layer of the feature extraction model. This layer preprocesses these patches, performing operations such as normalization and scaling to ensure a uniform format and range for subsequent modules.

[0044] The continuous spatial attention mechanism module uses parallel computing to process multiple local image blocks at the same time, converting the information of each local image block into a continuous signal representation. For example, the attention weights of each area in the image are determined by a linear combination of radial basis functions to highlight the key feature areas, namely the areas where fine lines and spots are located, and more effectively capture the associations between different areas in the image, especially the capture of detailed features such as facial fine lines and spots.

[0045] At the same time, because the continuous spatial attention mechanism module generates a large amount of feature data when processing images, the present invention sets a dynamic feature screening threshold to prevent excessive data from entering the subsequent processing stage. After the continuous spatial attention mechanism module outputs the feature data, the viscous memory module compares this feature data with the dynamic feature screening threshold. When the amount of feature data output exceeds the dynamic feature screening threshold, it prioritizes retaining feature data that is strongly related to facial fine lines and spots, while discarding relatively unimportant feature data. This prevents the viscous memory module and the entire model from experiencing problems such as slow processing speed, excessive memory usage, and even model crashes due to excessive data volume.

[0046] In this way, the final intermediate feature data matrix not only reduces the amount of data and improves the processing speed of the model, but also ensures that key features are retained.

[0047] As an example, the dynamic feature screening threshold is determined in the following manner: Calculating the texture complexity index of the three-dimensional facial skin image using a gray level co-occurrence matrix, and obtaining the image resolution of the three-dimensional facial skin image, The dynamic feature screening threshold is obtained by comprehensive calculation based on the texture complexity index and the image resolution.

[0048] Although the dynamic feature screening threshold can prevent excessive data from entering the subsequent processing stage and thus improve the processing speed of the model, if the dynamic feature screening threshold is set too low, the amount of data entering the subsequent processing steps will be insufficient, resulting in errors in the model's calculation results or a significant reduction in confidence.

[0049] To this end, the present invention extracts texture complexity index and image resolution from 3D facial skin images, and comprehensively calculates appropriate dynamic feature screening thresholds based on the texture complexity index and image resolution. The details are as follows: (1) The higher the texture complexity of a facial skin image, the richer the details of the skin surface. When using the ∞-former for feature extraction, more feature data will be generated. For example, a skin image with dense wrinkles and rough pores has a high texture complexity, and the feature extraction model will extract a large number of features related to wrinkles, pores, etc. Therefore, adjusting the feature screening threshold based on texture complexity can make the threshold match the feature richness of the image itself. When the texture complexity is high, the dynamic feature screening threshold is appropriately increased to avoid over-screening and loss of key features; when the texture complexity is low, the dynamic feature screening threshold is lowered to reduce redundant features and improve processing efficiency.

[0050] Before feeding the 3D facial skin image into the feature extraction model, the texture complexity of the 3D facial skin image is calculated using the Gray Level Co-occurrence Matrix (GLCM). The GLCM calculates the frequency of occurrence of specific grayscale values within a specific spatial relationship within the image, extracting texture feature parameters such as contrast, entropy, angular second moment, and inverse moment. These parameters are combined and weighted to produce a quantitative texture complexity metric. The weighting can be determined experimentally and empirically. For example, contrast and entropy are more important in reflecting texture complexity and can be given higher weights.

[0051] The calculation formula of texture complexity index can be: ,in, Contrast is used to reflect the clarity and grayscale contrast of image texture. The higher the contrast, the greater the grayscale difference between adjacent pixels in the image and the clearer the texture. Its calculation formula is: ; is the probability of occurrence of pixel pairs with grayscale values i and j in the grayscale co-occurrence matrix, and N is the number of grayscale levels of the image.

[0052] Entropy is used to measure the degree of disorder or amount of information in image texture. The larger the entropy value, the more complex and disordered the image texture is. The calculation formula is: .

[0053] It is the angular second moment (also called energy), which is used to reflect the uniformity of image grayscale distribution and texture coarseness. The larger the value, the more uniform the image grayscale distribution and the more regular the texture. The calculation formula is: .

[0054] It is the inverse moment (also called homogeneity), which is used to measure the local uniformity of image texture. The larger the value, the more uniform the image texture. The calculation formula is: , are the weights corresponding to contrast, entropy, angular second-order moment, and inverse moment, respectively, and satisfy .

[0055] (2) Image resolution directly affects the number of pixels in an image. The higher the resolution, the richer the information contained in the image, and the larger the amount of feature data generated during feature extraction. By incorporating resolution into the dynamic feature screening threshold calculation, the screening criteria can be dynamically adjusted according to the image data size. For high-resolution images, the dynamic feature screening threshold is appropriately increased to ensure that important features are retained when processing large amounts of data; for low-resolution images, the dynamic feature screening threshold is lowered to promptly eliminate possible noise and irrelevant features.

[0056] The resolution of the three-dimensional facial skin image is obtained, including the horizontal resolution and the vertical resolution, and the pixel scale information of the image is quantified by calculating the resolution product S=W*H.

[0057] After determining the texture complexity and image resolution, the first and second dynamic feature screening thresholds can be determined by matching the texture complexity and image resolution with their respective comparison tables. The final dynamic feature screening threshold can then be determined by weighted averaging. The comparison tables and their respective weighting coefficients can be derived through extensive testing and are not specifically limited.

[0058] By specifically adjusting the dynamic feature screening threshold, the present invention can accurately screen key features related to facial fine lines and spots based on the characteristics of different images, reducing unnecessary feature data processing and significantly improving the operational efficiency of the feature extraction model. When processing images with complex textures or high resolution, this method avoids the need for re-extracting features due to excessive screening, saving time. When processing simple images, it quickly eliminates redundant information, speeding up processing.

[0059] As an example, the facial skin condition data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model. The rating analysis model outputs facial fine line levels and facial spot levels, including: Connecting a facial skin database to the rating analysis model so that the rating analysis model performs self-training using facial skin data in the facial skin database; the facial skin data includes pre-configured facial skin condition data and grade labels, the grade labels including facial fine line grade labels and facial spot grade labels; After completing the self-training, the rating analysis model processes the facial skin condition data, classifies the facial fine lines level and facial spot level, and outputs them.

[0060] First, a connection is established between the rating analysis model and the facial skin database. Through appropriate interfaces and protocols, the rating analysis model can obtain the required facial skin data from the facial skin database.

[0061] The facial skin database stores pre-configured facial skin data, part of which is facial skin status data, which may be obtained through previous collection and processing, such as the number, length, depth, and direction of facial fine lines, and the size, shape, color, and distribution area of facial spots, among other detailed feature information; the other part is level labels, which include facial fine line level labels and facial spot level labels, which are used to indicate the level of the corresponding facial skin condition. For example, the facial fine line level can be divided into levels of 1-10, and the facial spot level is similarly set.

[0062] The rating analysis model uses this labeled facial skin data for self-training. During training, the rating analysis model continuously learns the mapping relationship between facial skin condition data and grade labels. After self-training, it can accurately predict the corresponding level of facial fine lines and facial pigmentation based on a set of facial skin condition data.

[0063] The output facial fine line level and facial pigmentation level are then matched with the formula database to generate a personalized beauty gummy formula for the user.

[0064] Among them, the rating analysis model can be constructed using classification algorithms such as SVM and random forest, or using neural network algorithms, without specific limitation.

[0065] An embodiment of the present invention further provides a cosmetic soft candy recipe recommendation system based on facial skin condition analysis, the system comprising a controller and a storage medium, wherein the storage medium stores a computer program, and the controller calls and executes the computer program to implement: The mobile phone camera is used to collect two-dimensional facial skin images of the user from multiple angles, and the two-dimensional facial skin images are synthesized into a three-dimensional facial skin image.

[0066] A feature extraction model based on ∞-former is used to identify facial fine line data and facial color spot data from the three-dimensional facial skin image to form facial skin status data; the facial skin status data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model, and the rating analysis model outputs the facial fine line level and the facial color spot level.

[0067] Based on the facial fine line level and the facial spot level, the recommended target formula is obtained by matching with a formula database. The target formula includes the dosage of collagen and antioxidants in the beauty gummy.

[0068] An embodiment of the present invention further provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements any of the aforementioned methods when executed by the processor.

[0069] An embodiment of the present invention further provides a computer storage medium storing a computer program that can be executed by a processor to implement any of the methods described above.

[0070] An embodiment of the present invention further provides a computer program product, which includes a computer program that can be executed by a processor to implement any of the methods described above.

[0071] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0072] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for recommending a beauty soft candy formula based on facial skin condition analysis, characterized by: The method comprises the following steps: Using a mobile phone camera to collect two-dimensional facial skin images of the user from multiple angles, and synthesizing the two-dimensional facial skin images into a three-dimensional facial skin image; Using an ∞-former-based feature extraction model, facial fine line data and facial color spot data are identified from the three-dimensional facial skin image to form facial skin condition data; the facial skin condition data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model, and the rating analysis model outputs a facial fine line level and a facial color spot level; Based on the facial fine line level and the facial spot level, the recommended target formula is obtained by matching with a formula database. The target formula includes the dosage of collagen and antioxidants in the beauty gummy.

2. The method for recommending a cosmetic soft candy formula based on facial skin condition analysis according to claim 1, wherein: The method of collecting a user's two-dimensional facial skin image from multiple angles using a mobile phone camera includes: Using a mobile phone camera to capture a two-dimensional facial skin image of the user's front face, a rough rating of the user's facial skin condition is performed based on the coarse features of facial fine lines and facial spots in the two-dimensional facial skin image, and a scanning and shooting frequency is determined based on the rough rating results; A formal scanning prompt signal is output, and two-dimensional facial skin images of the user at multiple angles are collected according to the scanning shooting frequency.

3. The method for recommending a cosmetic soft candy formula based on facial skin condition analysis according to claim 1, wherein: The feature extraction model includes an input layer, a continuous spatial attention mechanism module, a viscous memory module, a feature extraction layer, and an output layer; the facial fine line data and facial spot data are identified from the three-dimensional facial skin image using the ∞-former-based feature extraction model, including: Segmenting the three-dimensional facial skin image into a plurality of local image blocks, and inputting each of the local image blocks into the input layer; The continuous spatial attention mechanism module simultaneously performs continuous signal conversion and attention weight calculation on multiple local image blocks through parallel computing; and when the amount of feature data output by the continuous spatial attention mechanism module exceeds the dynamic feature screening threshold, the viscous memory module preferentially retains feature data strongly related to facial fine lines and spots to obtain an intermediate feature data matrix; The feature extraction layer uses a lightweight neural network structure to extract and separate facial fine line sub-data and facial color spot sub-data corresponding to each local image block from the intermediate feature data matrix; The output layer summarizes each of the facial fine line sub-data and each of the facial color spot sub-data, and integrates them to generate the facial fine line data and the facial color spot data.

4. The method for recommending a cosmetic soft candy formula based on facial skin condition analysis according to claim 3, wherein: The dynamic feature screening threshold is determined in the following manner: Calculating the texture complexity index of the three-dimensional facial skin image using a gray level co-occurrence matrix, and obtaining the image resolution of the three-dimensional facial skin image, The dynamic feature screening threshold is obtained by comprehensive calculation based on the texture complexity index and the image resolution.

5. The method for recommending a cosmetic soft candy formula based on facial skin condition analysis according to claim 4, characterized in that: The facial skin condition data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model. The rating analysis model outputs facial fine line levels and facial spot levels, including: Connecting a facial skin database to the rating analysis model so that the rating analysis model performs self-training using facial skin data in the facial skin database; the facial skin data includes pre-configured facial skin condition data and grade labels, the grade labels including facial fine line grade labels and facial spot grade labels; After completing the self-training, the rating analysis model processes the facial skin condition data, classifies the facial fine lines level and facial spot level, and outputs them.

6. A beauty soft candy formula recommendation system based on facial skin condition analysis, characterized in that: The system includes a controller and a storage medium, wherein the storage medium stores a computer program, and the controller calls and executes the computer program to implement: Using a mobile phone camera to collect two-dimensional facial skin images of the user from multiple angles, and synthesizing the two-dimensional facial skin images into a three-dimensional facial skin image; Using an ∞-former-based feature extraction model, facial fine line data and facial color spot data are identified from the three-dimensional facial skin image to form facial skin condition data; the facial skin condition data is input into a rating analysis model, and a facial skin database is connected to the rating analysis model, and the rating analysis model outputs a facial fine line level and a facial color spot level; Based on the facial fine line level and the facial spot level, the recommended target formula is obtained by matching with a formula database. The target formula includes the dosage of collagen and antioxidants in the beauty gummy.

7. A cosmetic soft candy recipe recommendation system based on facial skin condition analysis according to claim 6, characterized in that: The method of collecting a user's two-dimensional facial skin image from multiple angles using a mobile phone camera includes: Using a mobile phone camera to capture a two-dimensional facial skin image of the user's front face, a rough rating of the user's facial skin condition is performed based on the coarse features of facial fine lines and facial spots in the two-dimensional facial skin image, and a scanning and shooting frequency is determined based on the rough rating results; A formal scanning prompt signal is output, and two-dimensional facial skin images of the user at multiple angles are collected according to the scanning shooting frequency.

8. An electronic device, characterized in that: The electronic device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method according to any one of claims 1 to 5 when executed by the processor.

9. A computer storage medium, characterized in that: The computer storage medium stores a computer program that can be executed by a processor to implement the method according to any one of claims 1 to 5.

10. A computer program product, characterized in that: The computer program product comprises a computer program that can be executed by a processor to implement the method according to any one of claims 1 to 5.

Images