Cosmetic efficacy multi-index comprehensive evaluation system and platform

Abstract

The application provides a cosmetic efficacy multi-index comprehensive evaluation system and platform, belongs to the intelligent evaluation technical field, adopts differentiated component detection correction logic for different forms of cosmetics, and avoids detection errors caused by sampling and use loss; through image correction and pixel-level difference analysis, skin appearance changes are quantified, and a skin quality adapted appearance first evaluation is generated; a second evaluation of the essence of the formula is obtained by combining component standard comparison, a visual efficacy map is constructed by fusing two-dimensional evaluation data, multi-index fusion judgment is completed relying on feature aggregation and a neural network model, and a comprehensive evaluation result is output. The application realizes objective, accurate and traceable quantitative evaluation of skin care and basic makeup cosmetics.

Description

Technical Field

[0001] This invention relates to the field of intelligent evaluation technology, and in particular to a comprehensive evaluation system and platform for multiple indicators of cosmetic efficacy. Background Technology

[0002] With the standardization of the cosmetics industry, verifying the authenticity and effectiveness of cosmetic efficacy claims has become a core aspect of industry regulation and product development. Accurate, comprehensive, and objective efficacy evaluation technologies are crucial for ensuring cosmetic quality and regulating market order. Currently, the industry's efficacy evaluation methods for cosmetics are mainly divided into two categories: formula ingredient testing and human skin trial effect evaluation.

[0003] In existing technologies, some testing institutions still employ a simple, uniform sampling method, directly obtaining component data through microspectral analysis after batch sampling of cosmetic samples, without fully considering the spatial distribution heterogeneity of boxed cosmetics after opening and use. For boxed cosmetics such as liquid foundation and pressed powder, there are significant differences in the location attributes after the product is opened and used. The central area of ​​the box is used more frequently, with greater component loss and obvious changes in surface condition, while the edge area is used less frequently and the components remain intact. The sample properties of samples from different distribution locations are significantly different. Although regulations stipulate that powder products should be sampled at multiple points, the test data is not corrected for the surface condition after use, and it is still impossible to completely eliminate the testing errors caused by usage loss.

[0004] Therefore, this invention proposes a comprehensive evaluation system and platform for the efficacy of skincare and basic makeup cosmetics based on multiple indicators. Summary of the Invention

[0005] This invention provides a comprehensive evaluation system and platform for multiple indicators of cosmetic efficacy to solve the aforementioned technical problems.

[0006] This invention provides a comprehensive evaluation system for multiple indicators of cosmetic efficacy, comprising: The sample collection module is used to collect several partial samples of the cosmetic product to be tested as target samples, and to perform microspectral analysis on each target sample separately; The composition analysis module is used to obtain the basic components based on the microspectral analysis results of each target sample and the surface usage map of each target sample at its current position in the placement box when the sample properties of the cosmetic to be tested are related to the sample distribution location. When the sample properties are not related to the sample distribution location, the basic composition is obtained based on the microspectral analysis results of each sample. The acquisition module is used to acquire baseline skin images and response skin images of test subjects with different skin types based on the cosmetic to be tested, as image pairs, and at the same time, extract efficacy dimensions from the efficacy description of the cosmetic to be tested. The change analysis module is used to lock the change points of the response skin image in the corresponding image pair according to the baseline skin image and mark each candidate change point according to the change magnitude, in accordance with the efficacy dimension and skin type, to obtain a significant change layer. The first evaluation module is used to divide the significant change layers of all image pairs according to skin type and efficacy dimension, obtain the layer division set of different efficacy dimensions under the same skin type, and determine the first evaluation of the corresponding efficacy dimension based on the layer division set. The second evaluation module is used to compare the basic components with the standard components based on all efficacy dimensions to determine the second evaluation for each efficacy dimension. The graph construction module is used to construct a single-dimensional evaluation graph for the corresponding efficacy dimension based on all first evaluations of the same efficacy dimension under different skin types and in combination with the second evaluations of the same efficacy dimension. The comprehensive evaluation module is used to extract the single-dimensional evaluation center vector of each single-dimensional evaluation map and input it into a pre-trained neural network model to obtain the comprehensive evaluation result of the cosmetic product to be tested.

[0007] Preferably, the composition analysis module includes: An array construction unit is used to obtain the microspectral analysis results of each target sample when the sample attributes are related to the sample distribution location, and to extract the content of the same component at different current locations, construct a component content array, and obtain the mean and variance of the content of the corresponding component. The residual unit is used to compare and analyze the surface usage map at each current position with the standard surface map, determine the set of image changes and the interfering factors affecting each image change, and obtain the reaction state of each component at the corresponding current position and the residual content based on the reaction state. Then, based on the residual content array at each current position, the residual mean and residual variance of each component are determined. The comparison unit is used to perform a first comparison between the residual mean and the content mean of the same component, a second comparison between the residual variance and the content variance of the same component, and a third comparison between the residual content array and the component content array at the same current position, to obtain a three-dimensional array. The matching unit is used to match the three-dimensional array with the array-component comparison table to obtain the final content of the corresponding component, and to obtain the basic composition based on the final content of all components.

[0008] Preferably, the change analysis module includes: The traversal and correction unit is used to perform multi-scale local neighborhood traversal on the reference skin image, construct an adaptive baseline feature vector for each pixel based on the color distribution and texture features of the pixels in each neighborhood, and simultaneously perform illumination deviation correction and reflective noise filtering on the response skin image based on the grayscale histogram of the image pair to obtain the corrected response skin image. The difference calculation unit is used to perform multi-scale feature difference calculation on each pixel of the corrected response skin image based on the adaptive baseline feature vector to obtain the color difference component and texture difference component of the corresponding pixel. The point locking unit is used to perform preliminary threshold filtering on the color difference component and texture difference component of a single pixel, and then retain only pixels with continuous and abrupt feature difference distribution in the neighborhood as candidate change points to complete the change point locking. The weighted fusion unit is used to weight and fuse the color difference component and the texture difference component according to the weights corresponding to the efficacy dimension to obtain the change range. The grading unit is used to divide the change amplitude into multiple levels based on the change amplitude and the baseline fluctuation threshold of the skin region to which the candidate change point belongs, and generate corresponding salience level labels to complete the salience labeling. The mapping unit is used to map candidate change points with the salience level labels to a preset layer space to generate a salience change layer that is hierarchical according to salience level and associated with skin region attributes.

[0009] Preferably, the traversal and correction unit includes: The removal sub-unit is used to perform skin semantic segmentation on the baseline skin image and the response skin image and generate a skin pixel mask. After removing non-skin interference areas, a pure skin region image is obtained. The adaptive matching subunit is used to divide the response skin region image into multiple uniform sub-regions according to the illumination gradient, and adaptively match each sub-region with the reference local gray-level histogram corresponding to the pure skin region image to complete the local illumination deviation correction. The repair subunit is used to locate reflective areas based on abnormal peak values ​​of the grayscale histogram, combine neighborhood texture and skin color features to distinguish between the captured reflective noise and the real highlight areas of the skin, and use a texture consistency interpolation algorithm to repair the reflective noise areas while preserving the real highlight details, and output the corrected response skin image.

[0010] Preferably, the first evaluation module includes: The first curve plotting unit is used to regard each significant change layer in the layer division set as the first layer, and to perform point mapping matching between each first layer and the application site of the cosmetic to be tested to obtain the layer change set of the same site point. The change amplitude of each significant change point in the layer change set is sorted in descending order to obtain the first curve. The sub-curve plotting unit is used to filter the initial quantity of each significance level in the set of layer changes, and randomly sort all the change amplitudes under the same initial quantity to obtain sub-curves; The sorting unit is used to stitch together the sub-curves under the corresponding significance levels in descending order of significance level to obtain the second curve; The fitting unit is used to perform fitting processing on the first curve, the second curve and each sub-curve respectively to obtain the corresponding first fitting line, second fitting line and sub-fitting line; The value analysis unit is used to determine the effect reference value of the corresponding efficacy dimension based on the first intercept difference and the first slope difference between the first fitted line and the second fitted line, and the second intercept difference and the second slope difference between the third fitted line of the broken line obtained by splicing the sub-fitted lines and the first fitted line. The evaluation unit is used to determine the first evaluation of the corresponding efficacy dimension based on the reference value of the effect and the initial number of salience levels in the corresponding layer change set.

[0011] Preferably, the second evaluation module includes: The integration and differentiation unit is used to obtain the ingredient parameters for each efficacy dimension from the parameter database of the cosmetic to be tested, integrate all ingredient parameters to obtain standard components, and compare the basic components with the standard components one by one to obtain differentiated components. The segmentation unit is used to divide the differential components according to the efficacy dimension to obtain several sub-differential groups, and match them with the difference-evaluation comparison table to obtain the second evaluation of the corresponding efficacy dimension.

[0012] Preferably, the map construction module includes: The normalization unit is used to extract all first evaluation data corresponding to each skin type for the same efficacy dimension, and combine them with the corresponding skin type weight coefficient to obtain the normalized appearance evaluation quantitative value of the skin type. The matching correction unit is used to perform ingredient efficacy matching correction on the second evaluation of the same efficacy dimension based on the correlation mapping relationship between efficacy dimension and core cosmetic ingredients. The graph generation unit is used to fuse the normalized appearance evaluation quantification value with the corrected ingredient evaluation data to construct a multi-dimensional evaluation dataset that includes skin type compatibility, efficacy achievement, and ingredient matching. Based on the multi-dimensional evaluation dataset, gradient coordinate modeling is performed to generate a single-dimensional evaluation graph corresponding to the efficacy dimension.

[0013] Preferably, the comprehensive evaluation module includes: Local units are used to perform uniform grid segmentation on the single-dimensional evaluation map, extract the efficacy strength parameters, skin type compatibility parameters and ingredient matching parameters contained in each grid block, and form the local feature vector of each block. The aggregation unit is used to perform weighted correction on all local feature vectors based on the skin texture difference weight and ingredient matching correction weight in the map construction stage, and to perform aggregation calculation on all weighted local feature vectors through the iterative centroid convergence algorithm to solve the global centroid feature that fits the overall map feature distribution. The feature processing unit is used to normalize the global centroid features and generate a single-dimensional evaluation center vector corresponding to the efficacy dimension.

[0014] This invention provides a platform for implementing any of the methods described in the comprehensive evaluation system for multiple indicators of cosmetic efficacy.

[0015] Compared with the prior art, the beneficial effects of this application are: A differential component correction logic based on the correlation between sample attributes and distribution location was proposed. Combined with the multi-dimensional features of the surface usage map, the detection error caused by usage loss and chemical changes was corrected, thereby improving the accuracy of component detection in boxed cosmetics. A pixel-level quantitative system for skin appearance changes was established. The problem of secondary effect interference between different effects was solved by the efficacy cross-interference separation algorithm, and the evaluation accuracy was effectively improved compared with the existing technology. A three-dimensional independent evaluation map of skin type compatibility, efficacy achievement, and ingredient matching was constructed. Global features were extracted through an iterative centroid algorithm, and intelligent fusion and judgment of multiple indicators were achieved by combining neural networks. The evaluation results were highly consistent with third-party manual evaluations.

[0016] This system can be seamlessly integrated into an industrial internet platform to achieve cloud deployment, data interconnection, and multi-terminal sharing, supporting the implementation of automated evaluation of batch products in the cosmetics industry. At the same time, it introduces a strategy optimization mechanism for evaluation rules and weight parameters, which can dynamically adjust the evaluation logic to adapt to different product categories and skin types, greatly improving the system's versatility and adaptability.

[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of a comprehensive evaluation system for multiple indicators of cosmetic efficacy in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the implementation method of a comprehensive evaluation system for multiple indicators of cosmetic efficacy in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] This invention provides a comprehensive evaluation system for the efficacy of cosmetics based on multiple indicators, such as... Figure 1 As shown, it includes: The sample collection module is used to collect several partial samples of the cosmetic product to be tested as target samples, and to perform microspectral analysis on each target sample separately; The composition analysis module is used to obtain the basic components based on the microspectral analysis results of each target sample and the surface usage map of each target sample at its current position in the placement box when the sample properties of the cosmetic to be tested are related to the sample distribution location. When the sample properties are not related to the sample distribution location, the basic composition is obtained based on the microspectral analysis results of each sample. The acquisition module is used to acquire baseline skin images and response skin images of test subjects with different skin types based on the cosmetic to be tested, as image pairs, and at the same time, extract efficacy dimensions from the efficacy description of the cosmetic to be tested. The change analysis module is used to lock the change points of the response skin image in the corresponding image pair according to the baseline skin image and mark each candidate change point according to the change magnitude, in accordance with the efficacy dimension and skin type, to obtain a significant change layer. The first evaluation module is used to divide the significant change layers of all image pairs according to skin type and efficacy dimension, obtain the layer division set of different efficacy dimensions under the same skin type, and determine the first evaluation of the corresponding efficacy dimension based on the layer division set. The second evaluation module is used to compare the basic components with the standard components based on all efficacy dimensions to determine the second evaluation for each efficacy dimension. The graph construction module is used to construct a single-dimensional evaluation graph for the corresponding efficacy dimension based on all first evaluations of the same efficacy dimension under different skin types and in combination with the second evaluations of the same efficacy dimension. The comprehensive evaluation module is used to extract the single-dimensional evaluation center vector of each single-dimensional evaluation map and input it into a pre-trained neural network model to obtain the comprehensive evaluation result of the cosmetic product to be tested.

[0022] In this embodiment, the cosmetics to be tested refer to skincare and basic makeup cosmetics for which a comprehensive efficacy evaluation needs to be performed, including solid boxed makeup and liquid skincare products. The target samples are multiple equal-volume, independent local samples taken from the complete cosmetic body according to standardized sampling rules. Microspectral analysis uses high-precision material composition detection and analysis technology to qualitatively and quantitatively detect various chemical components, material structures, and component content ratios in the samples, accurately outputting the basic component data of the samples. For example, samples are taken from the central area, surrounding edges, and four corners of the foundation container, collecting a total of 8 target samples, and performing microspectral analysis on each sample, recording the component content data of samples at different locations; similarly, samples are taken from the upper, middle, and lower layers of the lotion bottle, collecting a total of 6 target samples, and performing microspectral analysis on each sample, recording the component data of evenly distributed samples.

[0023] In this embodiment, sample attributes refer to the physical and component states of the cosmetic sample, including detectable characteristics such as component content, viscosity, surface smoothness, and component loss. Sample distribution location refers to the specific location of the target sample in the original container of the cosmetic, such as the center or edge of boxed makeup, or the upper or lower layer of bottled skincare products. The surface usage image is a real-world image of the cosmetic surface collected using uniform parameters (such as light intensity 5000K, shooting angle 90°, and resolution 2048×2048), which can intuitively reflect the surface characteristics such as the degree of wear and tear, surface color, smoothness, unevenness, and loss at the sample location. The basic composition is the final determination of all effective ingredients contained in the cosmetic to be tested, the precise content of each ingredient, and the ratio of each ingredient after position error correction and data calibration.

[0024] Quantitative criteria for determining the correlation between sample attributes and distribution location: A sample attribute is considered to be correlated with its distribution location when any of the following conditions are met: Variance of component content at different sampling locations ; The wear difference in the surface image is ≥20%, where wear = (mean grayscale value of standard surface image - mean grayscale value of surface image after use) / mean grayscale value of standard surface image × 100%; for dark cosmetics (mean grayscale value of standard surface image < 50), grayscale inversion correction is used. To avoid abnormal negative abrasion levels, otherwise it is considered irrelevant. Based on statistical data from 1000 cosmetic products, the mean variance of ingredient content for boxed color cosmetics was 0.12, and the mean abrasion difference was 35%; the mean variance of ingredient content for fluid cosmetics was 0.008, and the mean abrasion difference was 3%. 0.05 and 20% are the optimal dividing thresholds for the two product categories. This represents the average grayscale value at the corresponding location in the standard surface image. The mean grayscale value of the corresponding position in the surface image after use is i=1,2,...,n, where n is the number of samples.

[0025] Skin type testers are volunteers covering various common skin types, including dry, oily, combination, and sensitive skin, used to simulate the product's effects on different groups. The baseline skin image is the original skin image captured before the tester used the tested cosmetic, under standardized shooting conditions and parameters, serving as a benchmark for skin condition comparison. The response skin image is the skin image captured after the tester used the tested cosmetic for a preset time, under the same shooting conditions and parameters as the baseline image, reflecting the skin's condition after product use. Image pairs are combinations of baseline and response skin images corresponding to the same tester and skin area. Efficacy dimensions are specific efficacy categories extracted based on cross-validation of cosmetic registration information from the National Medical Products Administration, product efficacy descriptions, and efficacy ingredient databases, eliminating false claims. For example, 20 testers each from dry, oily, combination, and sensitive skin types were selected, and baseline images of the T-zone and cheeks before use and response images after 28 days of use were collected for each person, forming four sets of massive image pairs corresponding to different skin types. If the product registration indicates moisturizing, oil control, brightening, soothing, and repairing effects, four core efficacy dimensions are ultimately extracted.

[0026] Change point locking is achieved by comparing image pixels and calculating feature differences to accurately locate pixel areas in a skin image that have changed in real condition compared to a baseline image, eliminating invalid and interfering areas. Change amplitude is the degree of skin change under the corresponding efficacy dimension, including quantitative values ​​such as changes in color depth, skin texture fineness, and pore condition. Salience labeling involves classifying each effective change point into a salience level based on the magnitude of the change and binding a corresponding level label. The salience change layer is a layered visual image layer that integrates all skin change points with salience level labels, providing a direct and quantitative display of skin efficacy improvements. For example, in the whitening efficacy dimension: it locks in pixel changes related to brightening facial skin tone and reducing dullness, calculates the change in skin tone brightness, and classifies it into three levels: significant brightening, slight brightening, and no change, generating a dedicated salience layer for whitening efficacy. Similarly, in the oil control efficacy dimension: it locks in changes in skin oil production and areas of oiliness, calculates the reduction in oil, classifies the oil control effect into levels, and generates a salience layer for oil control efficacy.

[0027] Efficacy Cross-Interference Separation Algorithm: An efficacy-feature contribution matrix is ​​established, and the contribution ratio of each efficacy dimension to color and texture features is pre-determined. For each candidate change point, its feature contribution value under each efficacy dimension is calculated, and efficacy dimensions with contribution values ​​below the threshold (0.1) are removed as interference. For example, the contribution of moisturizing efficacy to color features is 0.3, and to texture features is 0.7; the contribution of brightening efficacy to color features is 0.9, and to texture features is 0.1. When a pixel is detected to have a significant color change but a weak texture change, it is determined to be a brightening efficacy change, and secondary interference from moisturizing efficacy is removed.

[0028] The layer classification set is a collection of all significantly changed layers categorized and integrated according to the same skin type and the same efficacy dimension. The first evaluation is an apparent efficacy evaluation based on actual user feedback on human skin, quantitatively reflecting the actual effect of cosmetics on different skin types. For example, for the oil-control efficacy layer set for oily skin: the statistical layer shows that 82% of the points with highly significant oil-control changes account for a high percentage, indicating a significant reduction in average skin oil, ultimately generating a high-scoring first evaluation for oil-control efficacy on oily skin. For the oil-control efficacy layer set for dry skin, there are very few points with changes, resulting in a low-scoring first evaluation, accurately reflecting the differences in skin type suitability.

[0029] The standard ingredients are industry-standard and pre-stored in a database. They correspond to the standard active ingredient formulas and content threshold ranges for each cosmetic efficacy dimension, serving as the evaluation benchmark for ingredient compliance and efficacy support. The second evaluation is an essential evaluation of the formula based on the cosmetic's ingredient structure and active ingredient content. For example, the standard ingredients for moisturizing efficacy include hyaluronic acid, glycerin, and panthenol, with a preset content threshold of 0.2%-5%. If the tested cosmetic contains all three core ingredients and their content is within the standard range, the second evaluation for moisturizing efficacy is excellent. If panthenol is missing, or the content of the core ingredients is below the standard threshold, the second evaluation is unqualified or average.

[0030] A single-dimensional evaluation map is a multi-dimensional visual data map generated by integrating skin type compatibility, actual usage effect, and formula ingredient effect for a single efficacy dimension. For example, for whitening efficacy: it integrates the whitening appearance evaluation data of dry, oily, and combination skin types, corrects the ingredient matching evaluation of the core whitening ingredients (niacinamide, arbutin), and finally generates a single-dimensional evaluation map of whitening efficacy that includes skin type compatibility, whitening effect achievement, and ingredient matching.

[0031] The single-dimensional evaluation center vector is the core feature vector obtained by normalizing and aggregating the global features of the single-efficiency evaluation map.

[0032] The neural network model is an intelligent computing model pre-trained on an independent third-party human-evaluated dataset. It can autonomously perform weighted fusion and comprehensive analysis of multi-dimensional, multi-indicator data. Neural network model parameters: Model structure: 3-layer fully connected neural network; Input layer: Number of neurons = number of efficacy dimensions × 3, input is the single-dimensional evaluation center vector of each efficacy dimension; Hidden layer 1: 64 neurons, activation function ReLU; Hidden layer 2: 32 neurons, activation function ReLU; Output layer: 1 neuron, activation function Sigmoid, output comprehensive evaluation score (0-1). Training dataset: A paired dataset of human evaluation data from 10,000 cosmetic products by authoritative third-party organizations and the preprocessed data of this system, of which 80% is used for training, 10% for validation, and 10% for testing; Training parameters: Optimizer: Adam, learning rate 0.001; loss function: mean squared error (MSE); batch size: 32; training epochs: 100, early stopping strategy (stop if the validation set loss does not decrease for 5 consecutive epochs); Model performance: The coefficient of determination for the test set was 0.92, the mean absolute error (MAE) was 0.03, and the consistency with human evaluation reached 92%.

[0033] Multiply the 0-1 values ​​output by the neural network by 100 to obtain the comprehensive evaluation score (0-100 points), with the following levels: Excellent (≥85), Good (70-84), Satisfactory (60-69), and Weak (<60). The output also includes the score and level of each efficacy dimension, the optimal skin type, and the advantages and disadvantages of the formula.

[0034] In this embodiment, the comprehensive evaluation result is a final overall evaluation conclusion that integrates all efficacy dimensions, skin type suitability, ingredient ratios, and actual usage effects. It includes multi-dimensional results such as comprehensive score, efficacy level, and suitable skin type. For example, if the central vectors of the four efficacy functions—moisturizing, whitening, oil control, and repair—are input into a neural network model, the model integrates the skin type suitability effect of each efficacy, ingredient matching degree, and actual usage effect, and outputs a comprehensive efficacy score of 89 points after weighted calculation, corresponding to an excellent level, indicating a comprehensive evaluation result suitable for all skin types.

[0035] The beneficial effects of the above technical solution are as follows: By using differentiated sampling analysis logic, it effectively solves the problem that traditional detection technologies cannot be adapted to different categories of cosmetics such as boxed makeup and fluid skin care products. It accurately corrects the defects in ingredient detection caused by differences in application location. At the same time, it constructs a two-dimensional evaluation system for formula ingredient evaluation and human appearance effect evaluation. Through refined image analysis, multi-dimensional data map fusion, and neural network intelligent comprehensive judgment, it solves the problems of single dimension, strong subjectivity, fragmented data, and low accuracy in traditional cosmetic efficacy evaluation. It can comprehensively, objectively, and quantitatively complete the multi-indicator comprehensive efficacy evaluation of all categories of cosmetics, and the evaluation results are consistent with the true quality of the products.

[0036] This invention provides a comprehensive evaluation system for the efficacy of cosmetics based on multiple indicators. The component analysis module includes: An array construction unit is used to obtain the microspectral analysis results of each target sample when the sample attributes are related to the sample distribution location, and to extract the content of the same component at different current locations, construct a component content array, and obtain the mean and variance of the content of the corresponding component. The residual unit is used to compare and analyze the surface usage map at each current position with the standard surface map, determine the set of image changes and the interfering factors affecting each image change, and obtain the reaction state of each component at the corresponding current position and the residual content based on the reaction state. Then, based on the residual content array at each current position, the residual mean and residual variance of each component are determined. The comparison unit is used to perform a first comparison between the residual mean and the content mean of the same component, a second comparison between the residual variance and the content variance of the same component, and a third comparison between the residual content array and the component content array at the same current position, to obtain a three-dimensional array. The matching unit is used to match the three-dimensional array with the array-component comparison table to obtain the final content of the corresponding component, and to obtain the basic composition based on the final content of all components.

[0037] In this embodiment, interfering factors include uneven lighting, surface stains, and component oxidation; Reaction states include physical loss or chemical decomposition.

[0038] In this embodiment, the wear rate = (average gray value of standard surface image - average gray value of surface image after use) / average gray value of standard surface image × 100%.

[0039] Method for calculating residual content: Residual content of physical loss: ,in, The loss coefficient is 0.8 for powders and 0.6 for pastes. Wear degree; Chemical decomposition residue content: ,in, t is the component decomposition rate constant (determined through accelerated aging experiments), and t is the product opening time. Total residual content: .

[0040] Method for creating an array-component reference table: We collected 10,000 unopened cosmetic products from all categories and tested their standard ingredient content. .

[0041] Simulated usage experiments were conducted on each cosmetic product, and samples were taken from different locations to determine the content after use. And take pictures of the surface to calculate the wear degree. Color changes Degree of clumping (B), Degree of oxidation (O); Calculate the three-dimensional array for each cosmetic product. ,in: ,in, The residual mean, This represents the average content. ,in, For residual variance, For content variance; , which is the position difference array.

[0042] The three-dimensional array and correction coefficients are trained using the random forest algorithm. The mapping relationship is used to obtain an array-component lookup table and correction coefficients. The error rate is ≤2%.

[0043] For example, the sample to be tested was an oil-controlling powder compact from a certain brand that had been used for one month. Two samples were taken from the center, the middle, and the edge, for a total of six samples. Microspectral analysis yielded the following original talc content array: [82%, 83%, 90%, 91%, 95%, 96%]; Average content Content variance ; Surface wear characteristics: center wear 20%, color change 5%, clumping degree 10%; middle wear 5%, color change 1%, clumping degree 2%; edge wear 1%, color change 0%, clumping degree 0%. The calculated residual content array is: [97.7%, 98.2%, 93.8%, 94.1%, 95.8%, 96.0%]; residual mean Residual variance ; 3D array: , , [15.7%, 15.2%, 3.8%, 3.1%, 0.8%, 0.0%]; The correction coefficients are obtained from the matching array-component lookup table. ; Final talc content: .

[0044] The beneficial effects of the above technical solution are as follows: by constructing a multi-dimensional image change set that includes wear, color change, agglomeration degree and oxidation degree, and considering the physical loss and chemical decomposition of the components, and combining three-dimensional array statistical analysis and pre-trained random forest comparison table for component correction, the error of component detection in boxed cosmetics is reduced compared with the simplified linear correction method, and the accuracy of basic component analysis is greatly improved.

[0045] This invention provides a comprehensive evaluation system for the efficacy of cosmetics based on multiple indicators. The change analysis module includes: The traversal and correction unit is used to perform multi-scale local neighborhood traversal on the reference skin image, construct an adaptive baseline feature vector for each pixel based on the color distribution and texture features of the pixels in each neighborhood, and simultaneously perform illumination deviation correction and reflective noise filtering on the response skin image based on the grayscale histogram of the image pair to obtain the corrected response skin image. The difference calculation unit is used to perform multi-scale feature difference calculation on each pixel of the corrected response skin image based on the adaptive baseline feature vector to obtain the color difference component and texture difference component of the corresponding pixel. The point locking unit is used to perform preliminary threshold filtering on the color difference component and texture difference component of a single pixel, and then retain only pixels with continuous and abrupt feature difference distribution in the neighborhood as candidate change points to complete the change point locking. The weighted fusion unit is used to weight and fuse the color difference component and the texture difference component according to the weights corresponding to the efficacy dimension to obtain the change range. The grading unit is used to divide the change amplitude into multiple levels based on the change amplitude and the baseline fluctuation threshold of the skin region to which the candidate change point belongs, and generate corresponding salience level labels to complete the salience labeling. The mapping unit is used to map candidate change points with the salience level labels to a preset layer space to generate a salience change layer that is hierarchical according to salience level and associated with skin region attributes.

[0046] In this embodiment, the multi-scale local neighborhood uses three sizes: 3×3, 5×5, and 7×7, for each pixel. Extract the following features to construct an 8-dimensional adaptive baseline feature vector. Color features (4-dimensional): Mean values ​​of H, S, and V channels in the HSV color space, and mean value of R channel in the RGB color space, dimensionless. Texture features (4-dimensional): Mean value of LBP (Local Binary Pattern) in a 3×3 neighborhood, LBP variance in a 5×5 neighborhood, and gray-level co-occurrence matrix contrast and energy in a 7×7 neighborhood.

[0047] In this embodiment, the color difference component Texture difference component Calculation method: ,in, , , , This represents the normalized difference for the corresponding color channel; ,in, , , , This represents the normalized difference for the corresponding texture features.

[0048] The adaptive baseline feature vector is a set of exclusive feature parameters that are automatically adapted and generated based on the neighborhood features of each pixel in the benchmark skin image, serving as a baseline for comparison of the original skin state; illumination deviation correction corrects the image brightness deviation caused by uneven light intensity and angle during the shooting process; reflection noise filtering removes invalid interference information such as specular reflection and noise generated by the device during shooting.

[0049] Color difference component is the difference in pixel hue, brightness, and saturation between the response image and the reference image, corresponding to changes in effects such as whitening, brightening, and evening skin tone; texture difference component is the difference in skin texture coarseness, pore size, and wrinkle depth between the response image and the reference image, corresponding to changes in effects such as anti-wrinkle, skin smoothing, and repair. For example, in whitening effect detection: when the pixel brightness of dark areas on the face increases, the color difference component value is significant, while the texture difference component value is weak.

[0050] In this embodiment, the quantization rules for locking change points are as follows: Preliminary threshold filtering: Single pixel color difference component And texture difference components Pixels that are selected are retained, otherwise they are discarded. The threshold is determined based on the baseline fluctuation statistics of 1000 skin images.

[0051] Continuous change determination: Among the retained pixels, if at least 3 of its 8-neighborhood pixels also meet the threshold condition, it is determined to be a continuous change point and is used as a candidate change point; otherwise, it is determined to be a noise point and is removed. For example, filtering out invalid data such as minor noise differences in single pixels and occasional brightness fluctuations, and retaining pixels in areas of continuous change such as brightening of the cheeks, reduction of oil in the T-zone, and fading of fine lines as candidate change points.

[0052] In this embodiment, the weights corresponding to the efficacy dimension are preset differentiated weight ratios for different efficacies, as shown in Table 1: Table 1. Weighting List of All-Efficacy Dimensions

[0053] In this embodiment, the change range .

[0054] In this embodiment, the baseline fluctuation threshold and significance level classification are as follows: Baseline fluctuation thresholds for different skin areas (maximum natural variation in healthy skin under daily conditions): T-zone: 0.2; cheeks: 0.15; around the eyes: 0.1; nasolabial folds: 0.18; Significance rating criteria (combining the change range A and the baseline fluctuation threshold T): Significant improvement: A≥2T; Slight improvement: T≤A<2T; No significant change: A<T. For example, for oily skin, an oil control effect greater than 0.6 is considered significant oil control, 0.3-0.6 is considered slight oil control, and less than 0.3 is considered no significant change. Each change point is marked according to the range to distinguish the strength of the effect improvement.

[0055] In this embodiment, the preset layer space is a uniformly sized visualization layer carrier used to systematically display all skin change data.

[0056] Skin area attributes: The attributes of different facial skin areas such as T-zone, cheeks, eye area, and nasolabial folds are classified. For example, high-level oil control change points in the T-zone and slight brightening change points in the cheeks are displayed in separate layers, and the corresponding area attributes are associated to generate a layered, quantifiable and visualized efficacy change layer.

[0057] The beneficial effects of the above technical solution are as follows: by constructing pixel-level adaptive baseline feature vectors through multi-scale local neighborhood traversal, the quantitative calculation method of color difference components and texture difference components is clarified. Combined with the continuous change point locking rule and efficacy differentiation weight fusion, the fine quantification of skin appearance changes is realized, which solves the problem of insufficient overall feature comparison accuracy of existing technologies and improves the recognition accuracy of skin changes of different efficacy types.

[0058] This invention provides a comprehensive evaluation system for multiple indicators of cosmetic efficacy, wherein the traversal and correction unit includes: The removal sub-unit is used to perform skin semantic segmentation on the baseline skin image and the response skin image and generate a skin pixel mask. After removing non-skin interference areas, a pure skin region image is obtained. The adaptive matching subunit is used to divide the response skin region image into multiple uniform sub-regions according to the illumination gradient, and adaptively match each sub-region with the reference local gray-level histogram corresponding to the pure skin region image to complete the local illumination deviation correction. The repair subunit is used to locate reflective areas based on abnormal peak values ​​of the grayscale histogram, combine neighborhood texture and skin color features to distinguish between the captured reflective noise and the real highlight areas of the skin, and use a texture consistency interpolation algorithm to repair the reflective noise areas while preserving the real highlight details, and output the corrected response skin image.

[0059] In this embodiment, skin semantic segmentation is a segmentation process that distinguishes between skin regions and non-skin regions in an image using image recognition technology; the skin pixel mask is used to accurately lock the effective pixel region of the skin, and to block all non-effective detection areas, such as automatically removing interfering areas such as hair strands, glasses frames, background walls, and collars in the captured image, and retaining only the facial skin region for subsequent analysis.

[0060] In this embodiment, the illumination gradient is the gradient distribution of illumination intensity differences in different regions of the image; the local grayscale histogram is a statistical map of the distribution of pixel grayscale values ​​in a single sub-region, used to characterize the local brightness state. For example, for unevenly illuminated images where the side face is darker and the front face is brighter, the brightness is calibrated by region to make the illumination of the whole image uniform and conform to the illumination standard of the reference image.

[0061] In this embodiment, the abnormal peak value of the grayscale histogram is the abnormal numerical peak value formed by sudden bright pixels in the image, which is mostly shooting reflection noise. In this embodiment, the specific algorithm for restoring reflective noise is as follows: Reflective area location: In the grayscale histogram, areas with a grayscale value ≥240 and a pixel percentage ≥0.1% are identified as reflective areas.

[0062] Distinguishing between reflections and true highlights: Calculate the texture variance within the region. Texture variance < 0.01 indicates reflection noise, while texture variance ≥ 0.01 indicates true highlights.

[0063] Texture consistency interpolation repair: Using a bicubic interpolation algorithm, the texture and skin color features of the surrounding 8 pixels of the reflective area are used to fill and repair the reflective area. The peak signal-to-noise ratio (PSNR) of the repaired image is ≥35dB. For example, the noise of reflective white spots in the instant of shooting the bridge of the nose is removed, and the real oily highlight area of ​​the T-zone is preserved, which removes interference and preserves the real skin effect.

[0064] The beneficial effects of the above technical solution are: by eliminating non-skin interference areas through skin semantic segmentation, by using local grayscale histogram adaptive matching to eliminate the influence of uneven lighting, and by using texture variance to distinguish between shooting reflection noise and real skin highlight areas, the solution effectively removes image interference while preserving real skin function features such as oiliness and redness, providing high-quality basic image data for subsequent change analysis.

[0065] This invention provides a comprehensive evaluation system for multiple indicators of cosmetic efficacy, wherein the first evaluation module includes: The first curve plotting unit is used to regard each significant change layer in the layer division set as the first layer, and to perform point mapping matching between each first layer and the application site of the cosmetic to be tested to obtain the layer change set of the same site point. The change amplitude of each significant change point in the layer change set is sorted in descending order to obtain the first curve. The sub-curve plotting unit is used to filter the initial quantity of each significance level in the set of layer changes, and randomly sort all the change amplitudes under the same initial quantity to obtain sub-curves; The sorting unit is used to stitch together the sub-curves under the corresponding significance levels in descending order of significance level to obtain the second curve; The fitting unit is used to perform fitting processing on the first curve, the second curve and each sub-curve respectively to obtain the corresponding first fitting line, second fitting line and sub-fitting line; The value analysis unit is used to determine the effect reference value of the corresponding efficacy dimension based on the first intercept difference and the first slope difference between the first fitted line and the second fitted line, and the second intercept difference and the second slope difference between the third fitted line of the broken line obtained by splicing the sub-fitted lines and the first fitted line. The evaluation unit is used to determine the first evaluation of the corresponding efficacy dimension based on the reference value of the effect and the initial number of salience levels in the corresponding layer change set.

[0066] In this embodiment, the layer division set is a collection of all significant change layers pre-classified and archived according to the rule of a single skin type + a single efficacy dimension; the first layer is each independent skin significant change layer in the layer division set, with each layer corresponding to the visual change data of the skin after one cosmetic application; the target area is a preset fixed detection area of ​​the human face, including the T-zone, left and right cheeks, chin, eye area, nasolabial folds, and other common skin care detection areas; the same-site point mapping matching is based on the coordinates of the fixed parts of the human face, binding and aligning the pixel change points in the first layer with the corresponding site coordinates one by one, ensuring that the image change data accurately corresponds to the physical position of the human skin and eliminating positional misalignment; the layer change set is a summary collection of all change points and change amplitude data of the first layer corresponding to the same target area of ​​the face after the position matching is completed, with each site corresponding to an independent layer change set.

[0067] The first sorting is a numerical sorting operation performed on all valid change points within the layer change set, according to a uniform rule from largest to smallest, to obtain the first curve, which reflects the distribution pattern of the overall strength of a single effect under the corresponding skin type.

[0068] The initial number of significance levels is the original statistical total number of valid change points in a single layer change set, marked as significant improvement, slight improvement, no obvious change, etc.; the same initial number is the data condition that all change points have the same sample size and consistent statistical dimensions under the same significance level; the sub-curve is a subdivided trend curve drawn independently for single significance level data, used to individually characterize the distribution characteristics of the change magnitude under a certain effect level.

[0069] The significance level priority is a preset effect level ranking standard, fixed as: significant improvement > slight improvement > no obvious change; sub-curve splicing is a data fusion operation that connects the coordinate data of multiple independent sub-curves end to end according to priority order, integrating them into a complete continuous curve; the second curve is an integrated curve formed by splicing sub-curves of different levels according to priority, which can fully present the sample proportion and amplitude distribution characteristics of different effect levels. For example, according to the priority of significant improvement-slight improvement-no obvious change, the three oil control effect sub-curves are spliced ​​end to end in sequence, and the nodes are smoothly transitioned to generate the second curve; the first part of the curve corresponds to high effect sample data, the middle part corresponds to medium effect data, and the last part corresponds to ineffective data, clearly showing the hierarchical distribution of the overall effect.

[0070] In this embodiment, the first fitted line is a standard linear line obtained by fitting the first curve, representing the original trend of the overall efficacy before stratification; the second fitted line is a standard linear line obtained by fitting the second curve, representing the overall trend of the efficacy after stratification and integration; and the sub-fitted line is a standard linear line obtained by fitting a single independent sub-curve, representing the trend of the effect of a single significance level.

[0071] In this embodiment, the effect reference value V (0-100 points) is calculated using the following formula: ,in: This represents the difference in intercepts between the first and second fitted lines. This represents the difference in slope between the first and second fitted lines. This is the difference in intercepts between the third fitted line and the first fitted line; The slope difference between the third fitted line and the first fitted line; weighting coefficients: =0.3, =0.4, =0.1, =0.2, determined by regression analysis based on 5000 sets of evaluation data. For example, the first slope difference was calculated to be 0.04 and the first intercept difference was calculated to be 0.03. The third fitted line (slope 0.70, intercept 0.16) was obtained by splicing the sub-fitted straight line and fitting the broken line. The second slope difference was calculated to be 0.02 and the second intercept difference was calculated to be 0.01. Substituting the four types of differences into the weighted formula, the final reference value of the oil control effect for oily skin was 91.2 points. The higher the score, the better the actual apparent effect.

[0072] In this embodiment, the percentage distribution of the number of grades is the proportion of the initial number of change points for each grade of significant improvement, slight improvement, and no significant change to the total number of effective change points, reflecting the sample coverage of efficacy.

[0073] First evaluation score ,in, , , These represent the proportions of the points of change corresponding to the significance level out of the total number of points of change; First evaluation level classification: Excellent: ;good: ;qualified: Weak: For example, the oil control effect for oily skin: the reference value of the effect V=95.5 points, the proportion of samples with significant improvement is 82%, the proportion of slight improvement is 13%, and the proportion of no change is 5%. After comprehensive weighted scoring, the final comprehensive score is 89.5 points. According to the matching level rules, the first evaluation result is generated: excellent oil control effect, suitable for oily skin, and excellent overall appearance improvement effect.

[0074] The beneficial effects of the above technical solution are: by using multi-curve fitting difference analysis to simultaneously characterize the overall intensity and distribution characteristics of efficacy, the logical error of the original formula is corrected, and a comprehensive evaluation method of effect reference value + grade proportion weighting is adopted, so that the evaluation results can truly reflect the actual apparent efficacy of the product.

[0075] This invention provides a comprehensive evaluation system for multiple indicators of cosmetic efficacy, wherein the second evaluation module includes: The integration and differentiation unit is used to obtain the ingredient parameters for each efficacy dimension from the parameter database of the cosmetic to be tested, integrate all ingredient parameters to obtain standard components, and compare the basic components with the standard components one by one to obtain differentiated components. The segmentation unit is used to divide the differential components according to the efficacy dimension to obtain several sub-differential groups, and match them with the difference-evaluation comparison table to obtain the second evaluation of the corresponding efficacy dimension.

[0076] In this embodiment, the component parameters are the types of effective components, standard content ranges, and component ratio requirements corresponding to each efficacy, which are pre-stored in the database; the differential components are the set of abnormal components that are missing, insufficient in content, or unbalanced in ratio after comparing the tested components with the standard components. For example, the standard components for repair efficacy include panthenol and ceramide. If the product to be tested does not detect ceramide and the panthenol content is lower than the standard threshold, then the two types of components are included in the differential components.

[0077] In this embodiment, the sub-difference group is a set of single-effect differential ingredients obtained by splitting according to different efficacy dimensions such as moisturizing, whitening, and oil control.

[0078] The standard component database is based on the following sources: List of effective ingredients specified in the "Cosmetic Safety Technical Specifications (2022 Edition)"; Active ingredients listed in the International Dictionary of Cosmetic Ingredients (ICID); Statistics on the formula data of 10,000 registered cosmetic products; Studies on the content of active ingredients published in authoritative domestic and international journals; Taking moisturizing effects as an example, the standard components and content thresholds are as follows: core components: sodium hyaluronate (0.05%-2%), glycerin (2%-20%), panthenol (0.2%-5%); auxiliary components: ceramide (0.01%-1%), squalane (0.5%-10%).

[0079] In this embodiment, the specific scoring rules of the difference-evaluation checklist are shown in Table 2: Table 2 Difference-Evaluation Comparison Table

[0080] Second evaluation score (0-100 points) = 100 - total deductions, the grading is consistent with the first evaluation.

[0081] The beneficial effects of the above technical solution are: it establishes a graded quantitative deduction system based on the type of difference, transforms qualitative component evaluation into comparable quantitative scores, eliminates the subjectivity and arbitrariness in the component evaluation process, and realizes standardized comparison of the efficacy support of different product formulations.

[0082] This invention provides a comprehensive evaluation system for multiple indicators of cosmetic efficacy, wherein the spectrum construction module includes: The normalization unit is used to extract all first evaluation data corresponding to each skin type for the same efficacy dimension, and combine them with the corresponding skin type weight coefficient to obtain the normalized appearance evaluation quantitative value of the skin type. The matching correction unit is used to perform ingredient efficacy matching correction on the second evaluation of the same efficacy dimension based on the correlation mapping relationship between efficacy dimension and core cosmetic ingredients. The graph generation unit is used to fuse the normalized appearance evaluation quantification value with the corrected ingredient evaluation data to construct a multi-dimensional evaluation dataset that includes skin type compatibility, efficacy achievement, and ingredient matching. Based on the multi-dimensional evaluation dataset, gradient coordinate modeling is performed to generate a single-dimensional evaluation graph corresponding to the efficacy dimension.

[0083] Normalization is a calculation method that converts multiple sets of data of different magnitudes into data of a standard magnitude to eliminate sample bias; the quantitative value of appearance evaluation is a comprehensive appearance efficacy score of multiple skin types that can be directly compared after standardization.

[0084] The skin type weighting coefficients are determined based on the statistical distribution of skin types in the population: dry skin: 0.25; oily skin: 0.25; combination skin: 0.3; sensitive skin: 0.2.

[0085] Normalized appearance evaluation value of skin texture Using the min-max normalization method: ,in, Let m be the weighting coefficient for the m-th skin type. This is the first evaluation score corresponding to the skin type. , , The value range is from 0 to 1.

[0086] ,in, The standard deviation of the first evaluation score for different skin types. The average score of the first evaluation for different skin types; The value ranges from 0 to 1, and the larger the value, the more balanced the efficacy is across different skin types; Ingredient efficacy matching correction factor Where k is the number of core components, The efficacy weight of the core ingredient (core ingredient weight 1.0, auxiliary ingredient weight 0.3). For actual measured content, This is the median value of the standard content; Corrected component evaluation quantification value The value ranges from 0 to 1.

[0087] Establish a three-dimensional Cartesian coordinate system, X-axis (skin texture adaptation) = Y-axis (efficiency achievement) = Z-axis (component matching degree) = .

[0088] A continuous gradient surface is generated using a linear interpolation algorithm, with gradient color levels ranging from blue (0 points) to red (100 points).

[0089] The key data points are marked on the graph: the first evaluation score for each skin type, the ingredient matching score, and the overall score.

[0090] In this embodiment, the multidimensional evaluation dataset is a structured data set that integrates three types of indicators: skin type compatibility, efficacy achievement, and ingredient matching.

[0091] The beneficial effects of the above technical solution are: a three-dimensional independent Cartesian coordinate system of skin type compatibility, efficacy achievement, and ingredient matching is constructed, the newly added efficacy achievement index reflects the balance of the product's effect across different skin types, and combined with ingredient efficacy matching correction, the generated single-dimensional evaluation map can intuitively and comprehensively display the product's multi-dimensional evaluation results.

[0092] This invention provides a comprehensive evaluation system for multiple indicators of cosmetic efficacy, wherein the comprehensive evaluation module includes: Local units are used to perform uniform grid segmentation on the single-dimensional evaluation map, extract the efficacy strength parameters, skin type compatibility parameters and ingredient matching parameters contained in each grid block, and form the local feature vector of each block. The aggregation unit is used to perform weighted correction on all local feature vectors based on the skin texture difference weight and ingredient matching correction weight in the map construction stage, and to perform aggregation calculation on all weighted local feature vectors through the iterative centroid convergence algorithm to solve the global centroid feature that fits the overall map feature distribution. The feature processing unit is used to normalize the global centroid features and generate a single-dimensional evaluation center vector corresponding to the efficacy dimension.

[0093] In this embodiment, uniform grid segmentation is a standardized segmentation operation that equally divides the complete evaluation map into several grid blocks; the local feature vector is a feature array composed of three types of parameters: efficacy strength, skin type suitability, and ingredient matching contained in a single grid block; the differential weight includes two types of global correction parameters: skin type differential weight and ingredient matching correction weight; feature normalization is a processing operation that uniformly converts the global centroid features into a standard data range to adapt to the model input requirements.

[0094] The specific steps of the iterative centroid convergence algorithm are as follows: The single-dimensional evaluation map is evenly divided into 10×10 grid blocks, for a total of 100 blocks.

[0095] Extract the local feature vector of each block ,in, This represents the average value of the efficiency parameters within the block. The average value of parameters is adapted to skin type. This represents the mean of the component matching parameters.

[0096] Weighted correction of local feature vectors: ,in, , , The weight is determined based on expert scores and data regression.

[0097] Initialize global center of gravity It is the mean of all local feature vectors.

[0098] Iterative calculation: ,in, This is the bandwidth parameter.

[0099] Iteration termination condition: Alternatively, after 100 iterations or more, the global centroid feature G is finally obtained.

[0100] The beneficial effects of the above technical solution are: it corrects the weighting formula of local feature vectors, making the weights correspond one-to-one with the parameters, and aggregates and extracts the global centroid features of the map through the iterative centroid convergence algorithm. Compared with the simple global averaging method, it can more accurately reflect the overall distribution characteristics of the evaluation data, and provides reliable input data for the subsequent comprehensive judgment of the neural network.

[0101] This invention provides a platform for implementing a method for a comprehensive evaluation system of multiple indicators of cosmetic efficacy as described above, the method comprising: Step 1: Take several partial samples of the cosmetic product to be tested as target samples, and perform microspectral analysis on each target sample; Step 2: When the sample properties of the cosmetic to be tested are related to the sample distribution location, the basic components are obtained based on the microspectral analysis results of each target sample and the surface usage diagram of each target sample at its current position in the placement box. When the sample properties are not related to the sample distribution location, the basic composition is obtained based on the microspectral analysis results of each sample. Step 3: Obtain baseline skin images and response skin images of test subjects with different skin types based on the cosmetic to be tested, as image pairs. At the same time, extract efficacy dimensions from the efficacy description of the cosmetic to be tested. Step 4: According to the efficacy dimension and skin type, lock the change points of the response skin image in the corresponding image pair according to the baseline skin image and mark the significance of each candidate change point according to the change magnitude to obtain the significant change layer; Step 5: According to skin type and efficacy dimension, divide the significant change layer of all image pairs to obtain the layer division set of different efficacy dimensions under the same skin type, and determine the first evaluation of the corresponding efficacy dimension based on the layer division set. Step 6: Compare the basic components with the standard components based on all efficacy dimensions to determine the second evaluation for each efficacy dimension; Step 7: Based on all the first evaluations of the same efficacy dimension under different skin types, and combined with the second evaluations of the same efficacy dimension, construct a single-dimensional evaluation map for the corresponding efficacy dimension; Step 8: Extract the single-dimensional evaluation center vector of each single-dimensional evaluation map and input it into the pre-trained neural network model to obtain the comprehensive evaluation result of the cosmetic to be tested.

[0102] In this embodiment, the comprehensive evaluation process for a certain moisturizing face cream is as follows: Sampling and microspectral analysis: Two samples were taken from the top, middle and bottom layers of the face cream bottle, for a total of six samples. Microspectral analysis was used to obtain the content of each ingredient. Composition analysis: If the sample properties and distribution location are determined to be unrelated, the mean content of each component is calculated to obtain the basic composition. Image acquisition: 20 test subjects each of dry, oily, combination, and sensitive skin types were recruited to collect baseline images before use and response images after 28 days of use; Change analysis: The image is subjected to illumination correction and reflection repair to identify change points. The secondary brightening interference caused by moisturizing is removed by the efficacy cross-interference separation algorithm to generate a significant change layer. First evaluation: The first evaluation score for the moisturizing effect of each skin type was calculated as follows: 82 points (dry), 75 points (oily), 88 points (combination), and 78 points (sensitive). Second evaluation: Compared with the standard ingredients for moisturizing efficacy, a total of 5 points were deducted, and the second evaluation score was 95 points. Map construction: Normalized skin appearance evaluation: ; Efficacy achievement: ; Ingredient matching degree: ; A single-dimensional evaluation map of moisturizing efficacy was generated, with three-dimensional coordinates of (0.81, 0.936, 0.95). Overall evaluation: The central vector of moisturizing efficacy [0.81, 0.936, 0.95] was extracted and input into the neural network model, resulting in an overall score of 88 points, which is considered good and optimally suited for combination skin.

[0103] In this embodiment, the rules for handling abnormal situations are as follows: Abnormal ingredient content correction: When the corrected content exceeds 100%, the maximum content of the ingredient in the same type of product shall be taken as the final content. Image acquisition anomaly: If the test subject's facial expression changes cause the area to be misaligned by more than 5%, the image will be reacquired; Abnormal efficacy claims: When the efficacy claimed in the product manual is not reflected in the filing information and there is no corresponding functional ingredient to support it, that efficacy dimension will be removed. Neural network input anomalies: When the center vector of a certain efficacy dimension has missing values, the mean of similar products is used to fill the gaps.

[0104] The beneficial effects of the above technical solution are: it transforms the core technology of the system into a standardized end-to-end evaluation process, realizing full automation from sample collection, component analysis, image acquisition to comprehensive evaluation output, which greatly improves the efficiency of cosmetic efficacy evaluation, while ensuring the consistency of evaluation results for different batches through unified parameter control.

[0105] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A comprehensive evaluation system for the efficacy of cosmetics based on multiple indicators, characterized in that, include: The sample collection module is used to collect several partial samples of the cosmetic product to be tested as target samples, and to perform microspectral analysis on each target sample separately; The composition analysis module is used to obtain the basic components based on the microspectral analysis results of each target sample and the surface usage map of each target sample at its current position in the placement box when the sample properties of the cosmetic to be tested are related to the sample distribution location. When the sample properties are not related to the sample distribution location, the basic composition is obtained based on the microspectral analysis results of each sample. The acquisition module is used to acquire baseline skin images and response skin images of test subjects with different skin types based on the cosmetic to be tested, as image pairs, and at the same time, extract efficacy dimensions from the efficacy description of the cosmetic to be tested. The change analysis module is used to lock the change points of the response skin image in the corresponding image pair according to the baseline skin image and mark each candidate change point according to the change magnitude, in accordance with the efficacy dimension and skin type, to obtain a significant change layer. The first evaluation module is used to divide the significant change layers of all image pairs according to skin type and efficacy dimension, obtain the layer division set of different efficacy dimensions under the same skin type, and determine the first evaluation of the corresponding efficacy dimension based on the layer division set. The second evaluation module is used to compare the basic components with the standard components based on all efficacy dimensions to determine the second evaluation for each efficacy dimension. The graph construction module is used to construct a single-dimensional evaluation graph for the corresponding efficacy dimension based on all first evaluations of the same efficacy dimension under different skin types and in combination with the second evaluations of the same efficacy dimension. The comprehensive evaluation module is used to extract the single-dimensional evaluation center vector of each single-dimensional evaluation map and input it into a pre-trained neural network model to obtain the comprehensive evaluation result of the cosmetic product to be tested.

2. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 1, characterized in that, The composition analysis module includes: An array construction unit is used to obtain the microspectral analysis results of each target sample when the sample attributes are related to the sample distribution location, and to extract the content of the same component at different current locations, construct a component content array, and obtain the mean and variance of the content of the corresponding component. The residual unit is used to compare and analyze the surface usage map at each current position with the standard surface map, determine the set of image changes and the interfering factors affecting each image change, and obtain the reaction state of each component at the corresponding current position and the residual content based on the reaction state. Then, based on the residual content array at each current position, the residual mean and residual variance of each component are determined. The comparison unit is used to perform a first comparison between the residual mean and the content mean of the same component, a second comparison between the residual variance and the content variance of the same component, and a third comparison between the residual content array and the component content array at the same current position, to obtain a three-dimensional array. The matching unit is used to match the three-dimensional array with the array-component comparison table to obtain the final content of the corresponding component, and to obtain the basic composition based on the final content of all components.

3. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 1, characterized in that, The change analysis module includes: The traversal and correction unit is used to perform multi-scale local neighborhood traversal on the reference skin image, construct an adaptive baseline feature vector for each pixel based on the color distribution and texture features of the pixels in each neighborhood, and simultaneously perform illumination deviation correction and reflective noise filtering on the response skin image based on the grayscale histogram of the image pair to obtain the corrected response skin image. The difference calculation unit is used to perform multi-scale feature difference calculation on each pixel of the corrected response skin image based on the adaptive baseline feature vector to obtain the color difference component and texture difference component of the corresponding pixel. The point locking unit is used to perform preliminary threshold filtering on the color difference component and texture difference component of a single pixel, and then retain only pixels with continuous and abrupt feature difference distribution in the neighborhood as candidate change points to complete the change point locking. The weighted fusion unit is used to weight and fuse the color difference component and the texture difference component according to the weights corresponding to the efficacy dimension to obtain the change range. The grading unit is used to divide the change amplitude into multiple levels based on the change amplitude and the baseline fluctuation threshold of the skin region to which the candidate change point belongs, and generate corresponding salience level labels to complete the salience labeling. The mapping unit is used to map candidate change points with the salience level labels to a preset layer space to generate a salience change layer that is hierarchical according to salience level and associated with skin region attributes.

4. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 3, characterized in that, The traversal and correction unit includes: The removal sub-unit is used to perform skin semantic segmentation on the baseline skin image and the response skin image and generate a skin pixel mask. After removing non-skin interference areas, a pure skin region image is obtained. The adaptive matching subunit is used to divide the response skin region image into multiple uniform sub-regions according to the illumination gradient, and adaptively match each sub-region with the reference local gray-level histogram corresponding to the pure skin region image to complete the local illumination deviation correction. The repair subunit is used to locate reflective areas based on abnormal peak values ​​of the grayscale histogram, combine neighborhood texture and skin color features to distinguish between the captured reflective noise and the real highlight areas of the skin, and use a texture consistency interpolation algorithm to repair the reflective noise areas while preserving the real highlight details, and output the corrected response skin image.

5. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 1, characterized in that, The first evaluation module includes: The first curve plotting unit is used to regard each significant change layer in the layer division set as the first layer, and to perform point mapping matching between each first layer and the application site of the cosmetic to be tested to obtain the layer change set of the same site point. The change amplitude of each significant change point in the layer change set is sorted in descending order to obtain the first curve. The sub-curve plotting unit is used to filter the initial quantity of each significance level in the set of layer changes, and randomly sort all the change amplitudes under the same initial quantity to obtain sub-curves; The sorting unit is used to stitch together the sub-curves under the corresponding significance levels in descending order of significance level to obtain the second curve; The fitting unit is used to perform fitting processing on the first curve, the second curve and each sub-curve respectively to obtain the corresponding first fitting line, second fitting line and sub-fitting line; The value analysis unit is used to determine the effect reference value of the corresponding efficacy dimension based on the first intercept difference and the first slope difference between the first fitted line and the second fitted line, and the second intercept difference and the second slope difference between the third fitted line of the broken line obtained by splicing the sub-fitted lines and the first fitted line. The evaluation unit is used to determine the first evaluation of the corresponding efficacy dimension based on the reference value of the effect and the initial number of salience levels in the corresponding layer change set.

6. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 1, characterized in that, The second evaluation module includes: The integration and differentiation unit is used to obtain the ingredient parameters for each efficacy dimension from the parameter database of the cosmetic to be tested, integrate all ingredient parameters to obtain standard components, and compare the basic components with the standard components one by one to obtain differentiated components. The segmentation unit is used to divide the differential components according to the efficacy dimension to obtain several sub-differential groups, and match them with the difference-evaluation comparison table to obtain the second evaluation of the corresponding efficacy dimension.

7. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 1, characterized in that, The map construction module includes: The normalization unit is used to extract all first evaluation data corresponding to each skin type for the same efficacy dimension, and combine them with the corresponding skin type weight coefficient to obtain the normalized appearance evaluation quantitative value of the skin type. The matching correction unit is used to perform ingredient efficacy matching correction on the second evaluation of the same efficacy dimension based on the correlation mapping relationship between efficacy dimension and core cosmetic ingredients. The graph generation unit is used to fuse the normalized appearance evaluation quantification value with the corrected ingredient evaluation data to construct a multi-dimensional evaluation dataset that includes skin type compatibility, efficacy achievement, and ingredient matching. Based on the multi-dimensional evaluation dataset, gradient coordinate modeling is performed to generate a single-dimensional evaluation graph corresponding to the efficacy dimension.

8. The comprehensive evaluation system for multiple indicators of cosmetic efficacy according to claim 1, characterized in that, The comprehensive evaluation module includes: Local units are used to perform uniform grid segmentation on the single-dimensional evaluation map, extract the efficacy strength parameters, skin type compatibility parameters and ingredient matching parameters contained in each grid block, and form the local feature vector of each block. The aggregation unit is used to perform weighted correction on all local feature vectors based on the skin texture difference weight and ingredient matching correction weight in the map construction stage, and to perform aggregation calculation on all weighted local feature vectors through the iterative centroid convergence algorithm to solve the global centroid feature that fits the overall map feature distribution. The feature processing unit is used to normalize the global centroid features and generate a single-dimensional evaluation center vector corresponding to the efficacy dimension.

9. A platform, characterized in that, A method for implementing the comprehensive evaluation system for multiple indicators of cosmetic efficacy as described in any one of claims 1-8, the method comprising: Step 1: Take several partial samples of the cosmetic product to be tested as target samples, and perform microspectral analysis on each target sample; Step 2: When the sample properties of the cosmetic to be tested are related to the sample distribution location, the basic components are obtained based on the microspectral analysis results of each target sample and the surface usage diagram of each target sample at its current position in the placement box. When the sample properties are not related to the sample distribution location, the basic composition is obtained based on the microspectral analysis results of each sample. Step 3: Obtain baseline skin images and response skin images of test subjects with different skin types based on the cosmetic to be tested, as image pairs. At the same time, extract efficacy dimensions from the efficacy description of the cosmetic to be tested. Step 4: According to the efficacy dimension and skin type, lock the change points of the response skin image in the corresponding image pair according to the baseline skin image and mark the significance of each candidate change point according to the change magnitude to obtain the significant change layer; Step 5: According to skin type and efficacy dimension, divide the significant change layer of all image pairs to obtain the layer division set of different efficacy dimensions under the same skin type, and determine the first evaluation of the corresponding efficacy dimension based on the layer division set. Step 6: Compare the basic components with the standard components based on all efficacy dimensions to determine the second evaluation for each efficacy dimension; Step 7: Based on all the first evaluations of the same efficacy dimension under different skin types, and combined with the second evaluations of the same efficacy dimension, construct a single-dimensional evaluation map for the corresponding efficacy dimension; Step 8: Extract the single-dimensional evaluation center vector of each single-dimensional evaluation map and input it into the pre-trained neural network model to obtain the comprehensive evaluation result of the cosmetic to be tested.

Images