Facial aging grade evaluation method and system based on multi-dimensional data

Abstract

The invention relates to the field of medical data processing, in particular to a facial aging grade evaluation method and system based on multi-dimensional data, and the method comprises the steps: obtaining a reference age span D1 of a historical user in each dimension; obtaining a growth trend of a facial data growth sequence based on D1, and recording the growth trend as M1; for the face data distribution of all similar users between adjacent reference ages under each dimension, obtaining a reference face data span D2 of each dimension; recording the average difference between the face data of the current user and the face data of the same-age history as M2, wherein the difference between the face data of the current user and the face data of the same-age history is greater than D2; and combining M1 and M2 according to the difference between the noise of the facial data of all historical users when the facial data changes along with the age under the reference age span and the noise distributed by the facial data, and then evaluating the facial aging level of the current user. According to the invention, the reliability of facial aging evaluation results is ensured.

Description

Technical Field

[0001] The present invention relates to the field of medical data processing, and in particular to a method and system for assessing facial aging levels based on multi-dimensional data. Background Art

[0002] When assessing facial aging, it's often necessary to assess facial aging from multiple dimensions. For example, data on facial wrinkles, subcutaneous fat thickness, skin rebound speed, stretch recovery rate, stratum corneum water content, sebum secretion, and spot area are collected. Machine learning models are then used to assess facial aging based on these dimensions. Using multi-dimensional facial data can theoretically ensure objective and reliable assessment results. However, collecting multi-dimensional facial data requires specialized equipment. It's difficult to frequently use the equipment for aging assessment on the same user over a long period of time. Without long-term facial data, it's difficult to describe the evolution of aging. Furthermore, because the test results for different dimensions of facial data are affected differently by individual differences, such as different users' baseline physical conditions and recent lifestyle habits, the same user's lifestyle or physical condition can also vary over time. This results in varying errors and interference in the test results for a single facial data analysis across different dimensions. Therefore, directly using all the facial data test results for the current user for aging assessment fails to yield objective results. Summary of the Invention

[0003] To solve the above problems, the present invention provides a facial aging level assessment method and system based on multi-dimensional data.

[0004] The facial aging level assessment method and system based on multi-dimensional data of the present invention adopts the following technical solutions: One embodiment of the present invention provides a method for assessing facial aging level based on multi-dimensional data, the method comprising the following steps: When the variation of facial data of all historical users in each dimension with age in any age span is greater than the distribution range of facial data in each dimension in the age span, the age span is recorded as the reference age span D1 of each dimension; Several historical users with the most similar facial data to the current user at each age are recorded as similar users; Based on the difference between the current user's age and D1, several reference ages are set. The current user's facial data in each dimension and the average facial data of similar users at several reference ages constitute a facial data growth sequence, and the growth trend of the facial data growth sequence is recorded as M1. For the facial data distribution of all similar users between adjacent reference ages in each dimension, the distribution range of facial data with the same distribution in the facial data distribution of all adjacent reference ages is recorded as the reference facial data span D2 of each dimension. For historical users of the same age whose facial data differ from the current user by more than D2, the average difference between the facial data of the current user and the facial data of the historical users of the same age is recorded as M2; The difference between the noise of the facial data of all historical users as they change with age within the reference age span and the noise of the facial data growth sequence is recorded as the combined validity of each dimension. The combined validity is used to combine M1 and M2 to evaluate the facial aging level of the current user.

[0005] Preferably, when the variation of the facial data of all historical users in each dimension with age in any age span is greater than the distribution range of the facial data in each dimension in the age span, the age span is recorded as the reference age span of each dimension, and the specific steps included are as follows: The age difference between any two historical users is recorded as an age span; for any age span K, and for all ages of all historical users, any age is recorded as the initial age K0, and K+K0 is recorded as the ending age; in any dimension, the absolute value of the difference between the ending age and the average facial data of all historical users at the initial age is recorded as the change in facial data with age under age span K; the maximum difference in facial data of historical users in the age interval [K0, K0+K] is obtained and recorded as the distribution range of facial data of all historical users under age span K, and the difference between the change in facial data with age under age span K and the distribution range is recorded as the first difference of age span K; After all ages of all historical users are taken as initial ages, the mean of all first differences corresponding to the age span K is taken as the preferred index of the age span K. Among all age spans, the age span with the preferred index greater than 0 and the smallest value is obtained, which is recorded as the reference age span of each dimension.

[0006] Preferably, the step of setting a plurality of reference ages based on the difference between the current user age and D1 includes the following specific steps: The age of the current user is recorded as K1, and K1-D1, K1-2×D1, K1-3×D1, ..., K1-m×D1 are all used as reference ages; where m represents a preset integer. When there is a reference age that is greater than the minimum age of all historical users, the reference age is deleted.

[0007] Preferably, the facial data of the current user in each dimension and the average facial data of similar users at a certain reference age form a facial data growth sequence, which includes the following specific steps: For all similar users under any reference age, obtain the first mean of the facial data of all similar users in each dimension; sort the first means of all reference ages in ascending order of reference age to obtain the facial data growth sequence of the current user, and add the facial data of the current user in each dimension to the end of the facial data growth sequence.

[0008] Preferably, the facial data distribution of all similar users between adjacent reference ages in each dimension, wherein the distribution range of facial data with the same distribution in the facial data distribution of all adjacent reference ages is recorded as the reference facial data span D2 of each dimension, comprises the following specific steps: The facial data of all similar users between any two reference ages in each dimension constitute the facial data distribution of adjacent reference ages; for all facial data distributions of all adjacent reference ages, the centers of all facial data distributions are aligned, and the facial data in all facial data distributions after center alignment are linearly normalized and then mean shift clustering is performed to obtain all categories; the category containing the most facial data is taken as the target category, and the standard deviation of the facial data before normalization in the target category is recorded as the reference facial data span of each dimension, recorded as D2.

[0009] Preferably, the difference between the noise of the facial data of all historical users as they change with age within a reference age span and the noise of the facial data growth sequence is recorded as the combined effectiveness of each dimension, which includes the following specific steps: For all ages of all historical users, obtain the average facial data of all historical users in each dimension at each age; Starting from any age, a number of ages are sampled at equal intervals. The difference between the sampled ages is equal to the reference age span. The average facial data corresponding to the sampled ages constitute the first sequence. Obtain the noise distribution F1 of the first sequence, obtain the noise distribution F2 of the facial data growth sequence, and record (F1-F2) / F1 as the combined effectiveness of each dimension.

[0010] Preferably, after the combined effectiveness of each dimension is obtained, the specific steps included are as follows: Obtain the mean of the density of all categories other than the target category, and record the ratio of this mean to the density of the target category as the uncertainty coefficient of each dimension. Obtain a revised priority index for the reference age span of each dimension. The revised priority index is negatively correlated with the uncertainty coefficient and positively correlated with the combined effectiveness of each dimension. Record the reference age span when the revised priority index is the largest as the updated reference age span. The updated reference age span is used to re-acquire the combined validity of each dimension and the values ​​of M1 and M2. The facial aging level of the current user is evaluated by combining M1 and M2 using the re-acquired combined validity.

[0011] Preferably, the specific steps for obtaining the growth trend are as follows: After linear normalization of the facial data growth sequence, the element value at each position in the facial data growth sequence is fitted into a straight line using the least squares method, and the slope of the straight line is recorded as the growth trend.

[0012] Preferably, the step of utilizing the combined effectiveness to combine M1 and M2 to evaluate the facial aging level of the current user comprises the following specific steps: Normalize the combined effectiveness of all dimensions; obtain a combination coefficient, wherein when the normalized combined effectiveness is greater than a preset effective threshold, the combination coefficient is equal to 1; when the normalized combined effectiveness is less than or equal to the preset effective threshold, the combination coefficient is equal to 0; The growth trend and the relative difference with the same age are averaged and multiplied by the combination coefficient, which is recorded as the aging change characteristics of the current user in each dimension; the aging change characteristics of the current user are used to represent the result of the combination of the growth trend and the relative difference with the same age; the facial data of each dimension of the current user and the aging change characteristics obtained in each dimension are used as the facial aging detection results of each dimension, and the facial aging detection results of all dimensions are input into the machine learning model to obtain the facial aging level.

[0013] Another embodiment of the present invention provides a facial aging level assessment system based on multi-dimensional data, the system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, all steps of the above-mentioned facial aging level assessment method based on multi-dimensional data are executed.

[0014] The beneficial effects of the technical solution of the present invention are: The present invention analyzes the growth trend of the current user's facial data in each dimension based on an appropriate age span (i.e., a reference age span) to describe the changing characteristics of the current user's facial aging with age. This ensures that the changing trend of facial data with age is more obvious, avoids the changing trend being obscured by the differences and distinctions in the facial data of historical users, and contributes to a further reliable assessment of facial aging.

[0015] In addition, the reference facial data span D2 of each dimension is analyzed based on an appropriate age span (i.e., the reference age span) to ensure that the average difference in facial data of historical users of the same age can avoid the interference of noise distribution caused by individual differences among historical users.

[0016] Finally, combining growth trends with the average differences among historical users of the same age for aging assessment allows for simultaneous aging assessment based on both the horizontal temporal evolution of the current user's aging and the longitudinal distribution of individual aging across the same age group. This mitigates the unreliable assessment results caused by a lack of long-term test results, differences in physical condition over time for the same user, and individual differences between users. This invention combines the noise of all historical users' facial data as they change with age within a reference age span with the noise differences in the facial data distribution. This avoids the inaccurate or invalid growth trends and average differences among historical users of the same age analyzed based on the reference age span, further ensuring the reliability of aging assessments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flowchart of the steps of a facial aging level assessment method based on multi-dimensional data provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0019] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of the multi-dimensional data-based facial aging assessment method and system proposed by the present invention. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0020] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0021] The specific scheme of the facial aging level assessment method and system based on multi-dimensional data provided by the present invention is described in detail below with reference to the accompanying drawings.

[0022] Example 1:

[0023] See also Figure 1 , which shows a flowchart of a method for assessing facial aging level based on multi-dimensional data according to an embodiment of the present invention, the method comprises the following steps: Step S101: Collect user's facial data in several dimensions.

[0024] The dimensions described in this embodiment include: average wrinkle depth, wrinkle density, average wrinkle length, subcutaneous fat thickness, skin rebound speed, etc.

[0025] The average wrinkle depth, wrinkle density, and average wrinkle length can be obtained using image processing techniques; subcutaneous fat thickness can be determined using ultrasound imaging; and skin rebound velocity can be determined by applying negative pressure to the skin and measuring the deformation recovery time. The test results for each of these dimensions are recorded as facial data for that dimension.

[0026] Step S102: When the variation of facial data of all historical users in each dimension with age in any age span is greater than the distribution range of facial data in each dimension in the age span, the age span is recorded as the reference age span of each dimension.

[0027] Users whose facial data have been collected and aging assessments have been performed are recorded as historical users. Facial data of all dimensions of historical users are stored in the database. The age of each user is also recorded in the database, where the age is accurate to each month. In other embodiments, an age can be recorded every three months.

[0028] For any dimension and the facial data of all historical users in that dimension, average the facial data of all historical users of the same age to obtain the average facial data at each age. Obtain the age difference between any two historical users (i.e., the absolute value of the age difference). For all age differences of historical users (only one age difference is retained if the age difference is the same), each age difference is recorded as an age span.

[0029] When the variation of facial data of all historical users in each dimension with age in any age span is greater than the distribution range of facial data in each dimension in the age span, the age span is recorded as the reference age span of each dimension, denoted as D1.

[0030] The reference age span of each dimension represents the minimum age span in which the facial data of historical users show an obvious change pattern as they change with age.

[0031] As an example, when the variation of facial data of all historical users in each dimension with age in any age span is greater than the distribution range of facial data in each dimension in the age span, the age span is recorded as the reference age span of each dimension, which includes the following steps: For any age span K, and for all ages of all historical users, any age is recorded as the initial age K0, and K+K0 is recorded as the ending age; in any dimension, the absolute value of the difference between the average facial data of the ending age and the average facial data of the initial age is recorded as the change of facial data with age under the age span K; the facial data of all historical users in the age range [K0, K0+K] are obtained, and the difference between the maximum and minimum values ​​of these facial data is recorded as the distribution range of facial data of all historical users under the age span K; the difference between the change of facial data with age under the age span K and the distribution range of facial data of all historical users under the age span K is recorded as the first difference of the age span K.

[0032] Among them, if the first difference is greater than 0 or the larger the first difference is, the more obvious the changing trend of all facial data with age is, and this changing trend will not be masked by the differences and distinctions in the facial data of historical users. The reasons for the differences and distinctions in the facial data of all historical users are due to the individual differences of historical users. The individual differences include differences in the basic physiological conditions of historical users, differences in living habits and physical conditions before detecting facial data, and these differences cannot be used to describe the differences in aging conditions; for example, the differences in facial data of some historical users are not caused by different aging conditions, but are caused by the historical users' food, clothing, housing, transportation or basic skin diseases. Even the same historical user may have different facial data in two detections within a short period of time due to different physical conditions in a short period of time.

[0033] In particular, if there is no historical user at a certain age, a linear interpolation algorithm is used to interpolate the average facial data at that age using the average facial data corresponding to the ages on both sides of the age.

[0034] At this point, for any age of all historical users, using that age as the initial age yields the first difference of age span K. For all ages of all historical users, using these ages as the initial ages yields multiple first differences corresponding to age span K. The mean of all first differences is used as the preferred index for age span K. Among all age spans, the age span with the smallest preferred index greater than 0 is obtained and recorded as the reference age span for each dimension.

[0035] Step S103: Several historical users whose facial data are most similar to the current user at each age are recorded as similar users.

[0036] The current user refers to a user who has collected facial data of all dimensions but has not yet undergone aging assessment; for each user (including historical users and current users), the vector composed of facial data of all dimensions is recorded as the facial feature of each user.

[0037] Similar users are obtained based on the facial features of historical users and the facial features of the current user. As an example, the process specifically includes: The facial data of the same dimension in the facial features of all historical users and the current user are linearly normalized to obtain normalized facial features; the cosine similarity between the normalized facial features of each historical user and the normalized facial features of the current user is recorded as the similarity between each historical user and the current user.

[0038] Among all historical users at each age, obtain N0 historical users with the greatest similarity to the current user and record them as similar users of the current user at each age. In this embodiment, N0 represents half of the number of historical users at each age (rounded down).

[0039] Step S104: setting several reference ages based on the reference age span; constructing a facial data growth sequence based on the facial data of the current user in each dimension and the average facial data of similar users at the several reference ages; and obtaining a growth trend of the facial data growth sequence.

[0040] For any dimension, perform the following processing: The current user's age is denoted as K1, and K1-D1, K1-2×D1, K1-3×D1, ..., K1-m×D1 are all used as reference ages; where m is an integer, and in this embodiment, m is set to 10. Specifically, if a reference age exists that is greater than the minimum age of all historical users, that reference age is deleted.

[0041] Among them, the difference between adjacent reference ages is equal to the reference time. As mentioned above, the facial data of historical users with adjacent reference ages have an obvious change trend. This change trend will not be masked by the differences and distinctions in the facial data of all historical users between adjacent reference ages. It is helpful to analyze the change trend of facial data over a long period of time based on different reference ages, which is beneficial to describing the temporal evolution of facial aging.

[0042] Furthermore, for all similar users at any reference age, the facial data of similar users and the current user in all dimensions have a greater similarity. Compared with other historical users other than similar historical users, the facial aging process of similar users has a certain degree of similarity with the current user.

[0043] For all similar users with a given reference age, obtain the first mean of their facial data for that dimension. Then, sort the first means of all reference ages in ascending order of reference age to obtain a growing sequence of the current user's facial data. Add the current user's facial data for that dimension to the end of the growing sequence. The growing sequence describes how the current user's facial data changes with age.

[0044] The growth trend of the facial data growth sequence is obtained, denoted as M1. The growth trend is used to describe the growth of the current user's facial data as he or she ages. The above process of obtaining the growth trend avoids the problem that the facial aging changes of historical users are masked by individual differences of users when their facial aging evolves with age. This allows users who do not have long-term facial data collection to evaluate their aging changes.

[0045] As an example, a method for obtaining a growth trend of a facial data growth sequence includes: The facial data growth sequence is linearly normalized, and the element value at each position in the facial data growth sequence is fitted into a straight line using the least squares method. The slope of the straight line is recorded as the growth trend.

[0046] So far, this embodiment has obtained the growth trend of the current user's facial data in each dimension.

[0047] Step S105 : For all similar users between adjacent reference ages, obtain the reference facial data span of each dimension, and obtain the relative difference of the same age in each dimension based on the reference facial data span.

[0048] For any set of adjacent reference ages, the facial data of historical users at those ages exhibits a distinct trend of change. This trend is not masked by the differences and distinctions in the facial data of all historical users between those ages. These differences and distinctions manifest as noise in the facial data. For all historical users between those ages, individual differences (including lifestyle and physical condition) prevent their facial data from exhibiting distinct age-related evolution and changes. Instead, the noise distribution in the facial data caused by these individual differences is apparent. For different adjacent reference ages, the noise distribution caused by individual differences is identical or similar. For example, the noise distribution caused by individual differences between users K1-D1 and K1-2×D1 is identical or similar to the noise distribution caused by individual differences between K1-2×D1 and K1-3×D1. In other words, the noise distribution caused by individual differences is independent of age.

[0049] Based on this, the following processing is performed for any dimension: For all similar users between any two reference ages, their facial data constitutes the facial data distribution for adjacent reference ages. For all facial data distributions for adjacent reference ages, all facial data distributions are center-aligned. The facial data in all these center-aligned facial data distributions are linearly normalized and then clustered using mean-shift clustering to obtain all clusters.

[0050] In this embodiment, the bandwidth in the mean shift clustering is set to 0.15. In other embodiments, the mean shift clustering can also be replaced by a clustering algorithm based on a Gaussian mixture model.

[0051] In this embodiment, the category containing the most facial data is used as the target category. The facial data in the target category represents facial data with the same distribution pattern in different facial data distributions. These facial data describe the noise distribution caused by individual differences among historical users. The standard deviation of the facial data before normalization in the target category is recorded as the reference facial data span of each dimension, recorded as D2.

[0052] The reference facial data span is represented as follows: only when the difference in facial data between users is greater than D2 can it be determined that there is a clear difference in aging conditions between users. If the difference in facial data between users is less than or equal to D2, then even when the facial data between users are different, it cannot be determined whether there is a clear difference in aging conditions between users.

[0053] As an example, methods for center-aligning all facial data distributions include: For any facial data distribution, obtain the mean of all facial data in the facial data distribution, calculate the difference between all facial data in the facial data distribution and the mean, and obtain a decentralized facial data distribution. All decentralized facial data distributions are the distributions of all facial data after center alignment.

[0054] The purpose of center alignment is to eliminate the influence of age in the distribution of all facial data.

[0055] Furthermore, the following processing is performed for any dimension: Obtain historical users of the same age as the current user. In this embodiment, historical users whose ages are greater than K1-D1 / 2 and less than K1+D1 / 2 are recorded as historical users of the current user's age. Among all historical users of the same age, if the absolute difference between the facial data of any historical user of the same age and the facial data of the current user is greater than D2, then the historical user of the same age is recorded as a valid user.

[0056] The difference between the current user's facial data and the valid user's facial data is calculated respectively, and the ratio of the mean of all differences to the current user's facial data is recorded as the relative difference of the current user's facial data at the same age in this dimension, which is expressed as M2.

[0057] The above-mentioned valid user and the current user are in the same age group. The difference in facial data between the valid user and the current user will not be affected by individual differences, and will not affect the judgment of the facial aging difference between the valid user and the current user. Therefore, the obtained relative difference of the same age can reliably reflect the aging of the current user relative to other users of the same age.

[0058] Step S106: Obtain the combined effectiveness of each dimension based on the noise of the facial data growth sequence, and use the combined effectiveness to combine M1 and M2 to evaluate the facial aging level of the current user.

[0059] This embodiment simultaneously utilizes facial data from multiple dimensions to assess the current user's facial aging level, theoretically ensuring an objective and reliable assessment result. However, collecting facial data from multiple dimensions requires specialized equipment. For the same user, frequent and long-term use of the equipment for aging assessment is not feasible, and the user lacks long-term facial data assessment results. Furthermore, because the assessment results for different dimensions of facial data are affected differently by individual differences, such as different users' basic physical conditions and recent living habits, and the same user's living habits or physical condition may vary at different times, a single facial data assessment result can have varying errors and interference across different dimensions. Therefore, directly using the facial data assessment results from all dimensions of the current user for aging assessment will fail to yield an objective assessment result.

[0060] In this embodiment, each dimension is analyzed and processed to obtain the growth trend of facial data in each dimension and the relative difference among peers in each dimension. The growth trend of facial data can be used to estimate the growth and change of facial data of the current user over a long period of time; and the relative difference among peers in each dimension describes the difference in facial data of the current user relative to peers. This embodiment combines the growth trend of facial data in each dimension with the relative difference among peers in each dimension to perform aging assessment. It can simultaneously perform aging assessment based on the horizontal temporal aging evolution of the current user and the longitudinal aging distribution of peers, thus reducing the problem of unreliable assessment results caused by the lack of long-term test results for users, differences in physical condition of the same user at different times, and individual differences between different users.

[0061] For any dimension, when the growth trend of facial data in that dimension is combined with the relative difference of the same age in each dimension, the following processing is performed: The noise of facial data of all historical users as they change with age within a reference age span is obtained, and the difference between the noise of facial data of similar users as they change with age within the reference age span is recorded as the combined effectiveness of each dimension. The combined effectiveness of all dimensions is normalized. In this embodiment, softmax is used for normalization. The normalized combined effectiveness is used to combine the growth trend and the relative difference of the same age group for facial aging assessment.

[0062] The greater the combined effectiveness of each dimension, the less noise interference there is in the facial data of similar users based on the reference age span compared to the facial data of all historical users based on the reference age span. This means that the growth trend and relative differences among peers calculated based on the reference age span under this dimension are more reliable, and it is more important to pay attention to the combined results of the growth trend and relative differences among peers under this dimension. The smaller the combined effectiveness of each dimension, the greater the noise interference there is in the facial data of similar users based on the reference age span. The growth trend and relative differences among peers obtained based on the reference age span and the facial data of similar users may still fail to avoid the following problems: users lack long-term facial data detection results, and the detection results of different dimensions are affected differently by individual differences. The single facial data detection results of users have different errors and interferences in different dimensions, resulting in unreliable evaluation results.

[0063] As an example, the error of facial data of all historical users as they change with age under a reference age span is obtained, and the difference between the error of similar users as they change with age under the reference age span is recorded as the combined effectiveness of each dimension. The methods include: For all ages of all historical users, each age corresponds to an average facial data; Starting from any age, several ages are sampled at equal intervals. For example, starting from age k0, the ages sampled are k0, k0+D1, k0+2×D1, and so on. Several ages are sampled at equal intervals from all ages of all historical users.

[0064] A first sequence consisting of average facial data corresponding to the sampled ages is obtained.

[0065] Thus, taking any age as the starting point, a first sequence is obtained, and taking all ages as the starting points, multiple first sequences are obtained. It should be noted that if a first sequence is a subsequence of another first sequence, then the first sequence will also be deleted.

[0066] Obtain the noise distribution of each first sequence, and the mean of the noise distribution of all first sequences is recorded as F1; Obtain the noise distribution F2 of the facial data growth sequence (obtained in step S104 ), and record (F1 − F2) / F1 as the combined effectiveness of each dimension.

[0067] This concludes the present embodiment.

[0068] Example 2:

[0069] This embodiment provides a method for updating the reference age span of each dimension, specifically including: In step S105 of the first embodiment, multiple categories are obtained in each dimension. In addition to the target category, other categories can also represent the noise distribution caused by individual differences. That is, when there are multiple categories, it means that the noise distribution caused by individual differences is not unique and not deterministic.

[0070] Based on this, the mean density of all categories other than the target category is obtained. The ratio of this mean to the density of the target category is recorded as the uncertainty coefficient for each dimension, where the density of a category refers to the number of facial data within the category. A larger uncertainty coefficient indicates that there are other noise distributions caused by individual differences in addition to the target category, indicating that the uncertainty of the noise distribution caused by individual differences is high, resulting in an inability to accurately determine the noise distribution caused by individual differences. This is due to the error in the reference age span in Example 1.

[0071] Furthermore, the ratio of the normalized combined validity to the uncertainty coefficient of each dimension was recorded as the modified priority index of the reference age span.

[0072] For all age spans whose preferred indicators in step S102 of Example 1 are greater than 0, each age span is used as the reference age span of each dimension, and then steps S103 to S105 of Example 1 are executed to obtain the growth trend and relative difference of the same age for each dimension, as well as the normalized combined effectiveness of each dimension; then, the modified priority indicator of each age span is obtained according to the above steps of this example.

[0073] The largest age span of the revised priority indicator is used as the updated reference age span for each dimension.

[0074] Using the updated reference age span of each dimension, steps S103 to S106 of the first embodiment are executed to implement facial aging assessment of the current user.

[0075] In the above process, by further updating the reference age span, the accurate noise distribution caused by individual differences can be extracted, and the growth trend of each dimension and the relative difference of the same age can be combined with the greatest possible effectiveness, further ensuring the reliability of the assessment of facial aging.

[0076] In other embodiments, the ratio of the correction priority index obtained in Example 2 to lgG+1 can be recorded as the correction priority index of the reference age span in other embodiments. Then, in these embodiments, facial data evaluation is performed based on the correction priority index according to the method of Example 2. Where G represents the value of the reference age span, and lg represents the logarithm with base 10. The smaller G is, the larger the correction priority index is, which can ensure that the updated reference age span is as small as possible, so as to more accurately estimate the changes in the current user's facial data. The purpose of using lgG+1 as the denominator to calculate the ratio is, on the one hand, to avoid the denominator being 0, and on the other hand, to avoid the denominator being too large.

[0077] Example 3:

[0078] Step S106 of the first embodiment includes: obtaining the noise distribution of each first sequence.

[0079] As an optional example, the noise distribution of each first sequence is obtained, including the following method: The first sequence is Gaussian filtered using a Gaussian filter kernel of length 5 to obtain a filtered sequence. The absolute value of the difference between the mean of all elements in the first sequence and the mean of all elements in the filtered sequence is recorded as the noise distribution of the first sequence.

[0080] As a preferred example, obtaining the noise distribution of each first sequence includes the following method: All elements in the first sequence and all elements in the filtered sequence are linearly normalized. Each element in the normalized first sequence is subtracted from each element in the normalized filtered sequence and the absolute value is taken to obtain a difference sequence.

[0081] The standard deviation of all elements in the difference sequence is recorded as the noise distribution of the first sequence.

[0082] Step S106 of the first embodiment includes: combining the growth trend and the relative difference among peers.

[0083] As an example, combining growth trends and relative differences among peers, the methods include: A combination coefficient is obtained, wherein when the normalized combination effectiveness is greater than a preset effectiveness threshold, the combination coefficient is equal to 1, and when the normalized combination effectiveness is less than or equal to the preset effectiveness threshold, the combination coefficient is equal to 0. In this embodiment, the preset effectiveness threshold is set to 0.45.

[0084] The average of the growth trend and the relative difference between peers is multiplied by the combination coefficient to obtain the current user's aging change characteristic. The current user's aging change characteristic represents the result of combining the growth trend and the relative difference between peers.

[0085] In this example, when the combined validity is low, the aging change characteristics are all marked as 0, indicating that the growth trend and relative differences among peers are not referenced (that is, the growth trend and relative differences among peers are not referenced when performing aging assessment).

[0086] It should be noted that in this embodiment, both the growth trend and the relative difference among peers are dimensionless, and the average of the two values ​​is directly calculated.

[0087] As another example, combining growth trends and relative differences between peers, the following methods are included: The three-dimensional vector formed by the growth trend, relative difference between peers, and normalized combined effectiveness is used as the aging change feature of the current user. The aging change feature of the current user is used to represent the result of combining the growth trend and relative difference between peers.

[0088] In this example, the aging changes of different users can be finely distinguished based on the aging change characteristics.

[0089] Other examples of combining growth trends with relative differences between peers include: The average of the growth trend and relative age difference is multiplied by the normalized combined effectiveness, and this is recorded as the aging change feature for the current user. The aging change feature for the current user represents the result of combining the growth trend and relative age difference. This example avoids misclassifying the current user as severely aged when the growth trend and relative age difference are large, but the combined effectiveness of each dimension is low.

[0090] In the first embodiment, step S106 includes combining the growth trend and the relative difference of the same age using the normalized combined effectiveness, and using the combined effectiveness for facial aging assessment. As an example, the specific process includes: The facial data of the current user in each dimension and the aging change features obtained in each dimension are used as the facial aging detection results in each dimension.

[0091] The facial aging detection results of all dimensions are input into the machine learning model to obtain the facial aging grade.

[0092] The machine learning model of this embodiment is a fully connected neural network, and the training method of the fully connected neural network is: Half of all historical users are randomly selected as sample users (this portion of data is no longer considered historical users). For any sample user, this sample user is treated as the current user. According to the above embodiment, facial aging detection results for each dimension of this sample user are obtained. The facial aging detection results for all dimensions of all sample users are recorded as sample data. The sample data of all sample users constitute a dataset. Each sample data is artificially assigned an aging level label. In this embodiment, there are 10 aging levels, represented by 0, 0.1, 0.2, ..., and 0.9, respectively. Smaller values ​​indicate less severe aging. An aging level of 0 indicates no aging.

[0093] The fully connected neural network is trained using the data set. The loss function during training is the cross entropy loss function, and the neuron parameter updating method is the stochastic gradient descent algorithm. The fully connected neural network used in this embodiment has 4 intermediate layers, and each intermediate layer has 5 neurons.

[0094] The fully connected network and training method are well-known technologies and will not be described in detail in this embodiment.

[0095] It should be noted that this embodiment is applicable to scenarios with a sufficient number of historical users, for example, when the average number of historical users at each age is greater than 10, and a more reliable aging assessment result can be further obtained.

[0096] Specifically, if the average number of historical users of each age group is less than 10, the first embodiment is not performed. Instead, the facial data of the current user in each dimension is directly used as the facial aging detection result for each dimension. The facial aging detection results for all dimensions are then input into the machine learning model according to the method of this embodiment to obtain the facial aging grade.

[0097] Example 4:

[0098] As an example, the average wrinkle depth, wrinkle density, and average wrinkle length can be obtained using image processing technology, including the following steps: The depth camera is used to collect facial images of users (including historical users and current users) from the front, including color images and depth images of the face. A semantic segmentation network (such as the DeepLabV3 network) is used to obtain the forehead area and the area around the eyes of the face.

[0099] The Canny edge detection algorithm was used to detect edges in the forehead and periocular areas. Each edge was considered a wrinkle. The ratio of the total number of edge pixels in each region to the number of pixels within each region was used to calculate the wrinkle density within that region. The average length of all edges in each region was calculated, and the ratio of this average to the width of the facial image was used as the denominator to normalize the average wrinkle length.

[0100] The average grayscale value of all edge pixels in each region on the depth map is obtained, and the ratio of this average value to 225 is recorded as the average wrinkle depth of each region. The purpose of using 255 as the denominator to calculate the ratio is to normalize the average wrinkle depth.

[0101] In this embodiment, the wrinkle density of the forehead area and the wrinkle density of the eye area are weighted and summed to obtain facial data of the wrinkle density dimension, wherein the weights of the forehead area and the eye area in the weighted sum are 0.2 and 0.8 respectively.

[0102] In this embodiment, the average wrinkle length of the forehead area and the average wrinkle length of the eye area are weighted and summed to obtain facial data of the average wrinkle length dimension, wherein the weights of the forehead area and the eye area in the weighted sum are 0.3 and 0.7, respectively.

[0103] In this embodiment, the average wrinkle depth of the forehead area and the average wrinkle depth of the eye area are weighted and summed to obtain facial data of the dimension of average wrinkle depth, wherein the weights of the forehead area and the eye area in the weighted sum are 0.3 and 0.7, respectively.

[0104] As an example, subcutaneous fat thickness can be obtained by ultrasound imaging, including methods such as: An ultrasound device is placed close to the facial skin to capture an ultrasound image. The tester reads the thickness of the facial subcutaneous fat based on the ultrasound image, recorded as L (in millimeters). L / L0 is recorded as the relative thickness of the subcutaneous fat, where L0 represents the maximum subcutaneous fat and is used for normalization. In this embodiment, L0 = 50 mm.

[0105] As an example, the skin rebound speed can be obtained by applying negative pressure to the skin and measuring the deformation recovery time, including the following methods: A depth camera is used to capture a video of the user's face rebounding after being compressed, and the mean grayscale value P1 of the pressed area in the first frame of the depth image is obtained. The heat source is detected to filter out a frame of depth image when the facial deformation is recovered, and the mean grayscale value P2 of the pressed area in the depth image is obtained. The ratio of the absolute value of the difference between P1 and P2 and 255 is recorded as the facial data of the skin rebound speed dimension.

[0106] Embodiment 5:

[0107] This embodiment provides a facial aging level assessment system based on multi-dimensional data. The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, all the steps of the above embodiment are executed.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A facial aging level assessment method based on multi-dimensional data, characterized in that: The method comprises the following steps: When the variation of facial data of all historical users in each dimension with age in any age span is greater than the distribution range of facial data in each dimension in the age span, the age span is recorded as the reference age span D1 of each dimension; Several historical users with the most similar facial data to the current user at each age are recorded as similar users; Based on the difference between the current user's age and D1, several reference ages are set. The current user's facial data in each dimension and the average facial data of similar users at several reference ages constitute a facial data growth sequence, and the growth trend of the facial data growth sequence is recorded as M1. For the facial data distribution of all similar users between adjacent reference ages in each dimension, the distribution range of facial data with the same distribution in the facial data distribution of all adjacent reference ages is recorded as the reference facial data span D2 of each dimension. For historical users of the same age whose facial data differ from the current user by more than D2, the average difference between the facial data of the current user and the facial data of the historical users of the same age is recorded as M2; The difference between the noise of the facial data of all historical users as they change with age within the reference age span and the noise of the facial data growth sequence is recorded as the combined validity of each dimension. The combined validity is used to combine M1 and M2 to evaluate the facial aging level of the current user.

2. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: When the variation of facial data of all historical users in each dimension with age in any age span is greater than the distribution range of facial data in each dimension in the age span, the age span is recorded as the reference age span of each dimension, and the specific steps are as follows: The age difference between any two historical users is recorded as an age span. For any age span K, and for all ages of all historical users, any age is recorded as the initial age K0, and K+K0 is recorded as the ending age. In any dimension, the absolute value of the difference between the ending age and the average facial data of all historical users at the initial age is recorded as the change in facial data with age under age span K. Obtain the maximum difference in facial data of all historical users within the age range [K0, K0+K], record this as the distribution range of facial data of all historical users within age span K, and record the difference between the change in facial data with age within age span K and the distribution range as the first difference for age span K. After all ages of all historical users are taken as initial ages, the mean of all first differences corresponding to the age span K is taken as the preferred index of the age span K. Among all age spans, the age span with the preferred index greater than 0 and the smallest value is obtained, which is recorded as the reference age span of each dimension.

3. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: The specific steps of setting a plurality of reference ages based on the difference between the current user age and D1 are as follows: The current user's age is recorded as K1, and K1-D1, K1-2×D1, K1-3×D1, ..., K1-m×D1 are all used as reference ages; Where m represents a preset integer. When there is a reference age that is greater than the minimum age of all historical users, the reference age is deleted.

4. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: The facial data of the current user in each dimension and the average facial data of similar users at a certain reference age constitute a facial data growth sequence, which includes the following specific steps: For all similar users under any reference age, obtain the first mean of the facial data of all similar users in each dimension; sort the first means of all reference ages in ascending order of reference age to obtain the facial data growth sequence of the current user, and add the facial data of the current user in each dimension to the end of the facial data growth sequence.

5. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: For the facial data distribution of all similar users between adjacent reference ages in each dimension, the distribution range of facial data with the same distribution in the facial data distribution of all adjacent reference ages is recorded as the reference facial data span D2 of each dimension, which includes the following specific steps: The facial data of all similar users between any two reference ages in each dimension constitute the facial data distribution of adjacent reference ages; for all facial data distributions of all adjacent reference ages, the centers of all facial data distributions are aligned, and the facial data in all facial data distributions after center alignment are linearly normalized and then mean shift clustering is performed to obtain all categories; the category containing the most facial data is taken as the target category, and the standard deviation of the facial data before normalization in the target category is recorded as the reference facial data span of each dimension, recorded as D2.

6. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: The difference between the noise of the facial data of all historical users as they change with age under the reference age span and the noise of the facial data growth sequence is recorded as the combined effectiveness of each dimension, including the following specific steps: For all ages of all historical users, obtain the average facial data of all historical users in each dimension at each age; Starting from any age, a number of ages are sampled at equal intervals. The difference between the sampled ages is equal to the reference age span. The average facial data corresponding to the sampled ages constitute the first sequence. Obtain the noise distribution F1 of the first sequence, obtain the noise distribution F2 of the facial data growth sequence, and record (F1-F2) / F1 as the combined effectiveness of each dimension.

7. The facial aging level assessment method based on multi-dimensional data according to claim 5, characterized in that: After the combined effectiveness of each dimension is obtained, the specific steps involved are as follows: Obtain the mean of the density of all categories other than the target category, and record the ratio of this mean to the density of the target category as the uncertainty coefficient of each dimension. Obtain a modified priority index of the reference age span of each dimension, wherein the modified priority index is negatively correlated with the uncertainty coefficient and positively correlated with the combined effectiveness of each dimension. The reference age span when the revised priority index is the largest is recorded as the updated reference age span; The updated reference age span is used to re-acquire the combined validity of each dimension and the values ​​of M1 and M2. The facial aging level of the current user is evaluated by combining M1 and M2 using the re-acquired combined validity.

8. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: The specific steps for obtaining the growth trend are as follows: After linear normalization of the facial data growth sequence, the element value at each position in the facial data growth sequence is fitted into a straight line using the least squares method, and the slope of the straight line is recorded as the growth trend.

9. The facial aging level assessment method based on multi-dimensional data according to claim 1, characterized in that: The method of utilizing the combined effectiveness to combine M1 and M2 to evaluate the facial aging level of the current user includes the following specific steps: Normalize the combined effectiveness of all dimensions; obtain a combination coefficient, wherein when the normalized combined effectiveness is greater than a preset effective threshold, the combination coefficient is equal to 1; when the normalized combined effectiveness is less than or equal to the preset effective threshold, the combination coefficient is equal to 0; The growth trend and the relative difference with the same age are averaged and multiplied by the combination coefficient, which is recorded as the aging change characteristics of the current user in each dimension; the aging change characteristics of the current user are used to represent the result of the combination of the growth trend and the relative difference with the same age; the facial data of each dimension of the current user and the aging change characteristics obtained in each dimension are used as the facial aging detection results of each dimension, and the facial aging detection results of all dimensions are input into the machine learning model to obtain the facial aging level.

10. A facial aging assessment system based on multi-dimensional data, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the processor runs the computer program, all steps of the facial aging level assessment method based on multi-dimensional data according to any one of claims 1 to 9 are executed.

Images