Wang Sijing,XIA Qianru,CHEN Gang,YU Jing

Shanghai Yongxi Information Technology Co,Ltd.,China

Abstract To investigate the apparent age of Chinese cosmetic consumers and its influencing factors. The subjects’ skin conditions in all dimensions were collected using professional instruments and clinical expert assessment,subjects’ lifestyles,skin care and makeup habits were obtained through questionnaire.The apparent age of the subjects was obtained based on the visual perception of photos judged by observers and then averaged.The association between apparent age and skin characteristics,the association between the difference between apparent age and actual age and the subjects' lifestyle and its difference among cities were investigated.The results showed that apparent age had a high correlation with skin tone and severity of skin problems.The model of multiple regression analysis obtained a high resolution (R2=0.704).The use of skin care products may help to delay the apparent aging of the skin.The results of the study have some guiding significance for the development of anti-aging products and the evaluation of anti-aging efficacy and is informative for lifestyle choices to maintain youthfulness.

Key words pparent age;cosmetic consumer;lifestyle;multi-center

Facial apparent Age,also known as perceived age,refers to the age a person appears to be,and is a visual judgment based on the condition of one’s facial skin,not one’s true age.People with lower apparent age tend to have better skin.

The existence of large public databases allows for relatively accurate estimation of apparent age from individual face images by software algorithms.Agustsson et al[1].developed a face depth residual regression method for predicting apparent age,which can map predicted apparent age to true age by onedimensional support vector regression.Li et al[2].used an improved VGGnet neural network model to train a dataset of 50,000 face images to improve the apparent age prediction accuracy,and the study also tested the effects of facial plastic surgery and makeup,which showed that different plastic surgery and makeup had varying degrees of impact on the facial apparent age.In the patented invention of Cui et al.[3],they collected facial photos of multiple subjects,and performed linear regression analysis on the real age of the subjects and the distance and angle of their facial lines,found that facial lines had a significant effect on apparent age.Flament et al.[4]discovered that the severity of skin problems such as wrinkles,sagging and hyperpigmentation were associated with the assessment of apparent age through the evaluation of 28 facial signs of aging atlas.Guyuron et al[5].administered a comprehensive questionnaire to 186 American identical twins and found that longer duration of smoking and UV exposure were associated with greater perceived age;facial wrinkles were more pronounced in smokers;and pigmentation was less pronounced in the higher weight twins.Mayes et al[6].collected facial photographs of 250 Chinese women aged 25 to 70 years and used a questionnaire to record their lifestyle and health information,and quantified the variables associated with perceived facial age through bivariate analysis and multiple linear regression identification,concluding that factors such as education level,number of family members,exposure to sunlight,and whether or not they used face creams and cleansers affected the subjects’ apparent age.Goodman et al.[7]surveyed 3,267 18 to 75-year-olds from the United States,Australia,Canada,and the United Kingdom women’s ratings of their levels of aging and their dietary habits.After linear regression analysis,it was found that smoking and alcohol consumption significantly increased facial apparent age.

Existing studies on deep learning algorithms for estimating apparent age based on facial images are relatively mature and accurate.Studies on the association between apparent age and skin condition in Caucasian and wider age ranges,and the effect of lifestyle on apparent age have also been reported.However,there is still a need for further research on the association between apparent age and the skin characteristics,the effect of lifestyle on apparent age,and inter-city differences in China’s main cosmetic consumer group.In this study,the mean values of apparent age of 118 subjects by 24 observers were used as dependent variables,37 factors such as pigmentation and texture assessed by experts and instruments were used as independent variables,and the correlation analysis by Pearson coefficient and linear regression analysis by stepwise method (analysis software spssau)were used to obtain the assessment model of apparent age and further investigate the effect of lifestyle on the apparent age of facial skin and the similarities and differences between the apparent and true ages across the cities studied.This will be a guideline for the development of anti-aging cosmetics and the study of cosmetic efficacy,especially anti-aging or agingconcealing efficacy.

1 The experiment

1.1 Instrument

Visia-7 (Canfield Scientific,Inc,.USA);Miniscan EZ spectrophotometer (HunterLab Associates Laboratory,Inc.,USA);Corneometer(Courage&Khazaka Electronic,Inc.,GER),Cutometer (Courage&Khazaka Electronic,Inc.,GER);Sebumeter (Courage&Khazaka Electronic,Inc.,GER)

1.2 Subject

Considering different geographic environments,population structures and economic development levels,118 female cosmetic consumers aged 18 to 45 years old with healthy skin in different cities were selected as study subjects,including Beijing(18),Shenyang (17),Shanghai (16),Wuhan (16),Guangzhou (14),Chengdu (17),Xi’an (20).Subjects who were pregnant,lactating,or had skin diseases (e.g.,dermatitis,psoriasis,eczema,etc.) were excluded.

Female cosmetic consumers aged 18 to 35 years old with healthy skin free of diseases and sensitive vision were recruited as observers for this study.Those with color blindness,color weakness,corrected visual acuity below 0.8 and other obvious visual impairment eye diseases were excluded.

1.3 Experimental method

1.3.1 Photo and skin feature data collection

All subjects cleansed their face at home with water and remained bare skin for the visit.Upon arrival,they first signed an informed consent form,followed by priority completion of sebum content testing (Sebumeter,both sides of the forehead).Subjects sat for 20 min at constant temperature and humidity room after cleansing their faces with water.The skin problems were then clinically assessed by experts (factors including: severity of acne,severity of blackheads,severity of whiteheads,severity of acne marks,severity of pitted acne scars,severity of dark circles,severity of undereye puffiness,severity of tear troughs,severity of undereye fine lines,severity of crow's feet,severity of forehead lines,severity of nasolabial folds,severity of enlarged pores,severity of local dullness,severity of yellowness,severity of pigmentation,severity of redness,severity of uneven skin tone) and objective assessment by instrument(factors including: Taking standard photographs of subjects using Visia-7 and analyzing the images for spots,forehead lines,texture,pores,UV spots,brown spots,red area,porphyrin,and fine lines around the eyes;using Miniscan EZ to test the skin tone of the subjects,the test area were forehead,both cheeks,both jaws,chin;using Corneometer to test the water content of the subjects’ skin,the test area was both cheeks,3 times each to take the average value;using Cutometer to assess the elasticity and firmness of the subjects’skin,the test area was both cheeks,1 time each).

After the photos and data collection were completed,photo-based apparent age assessment was performed: the frontal standard photos of 118 subjects were shown to 24 observers in turn for observation,and the observers predicted the apparent age of the 118 photos by observing only the skin condition of the subjects without considering the facial features.The apparent age of the above 118 subjects was obtained by averaging the numbers given by 24 observers.

1.3.2 Lifestyle questionnaire collection

In this study,information related to personal lifestyle such as household income,dietary habits(cooking,eating sweets,drinking milk,etc.),habit of staying up late,habit of smoking (frequency),and habit of using skincare products of 118 subjects were recorded by means of a questionnaire.

1.4 Data statistics and analysis methods

1.4.1 Kendall coefficient of cooperation

Wheremis the number of evaluators,nis the number of individuals,and is the sum of ranks of the i-th individual.Sis the sum of squares of deviations of the rank-sum for each individual from the mean rank:S=≤W≤1.

When the evaluations are consistent,Sis at its maximum,W=1.Therefore,Kendall's Coefficient of Concordance is used to study the degree of consistency amongmevaluators in their evaluations ofnindividuals.

1.4.2 Pearson correlation coefficient

For variables X and Y with n samples (x1,y1),(x2,y2),…,(xn,yn),the Pearson correlation coefficient

r is used to measure the correlation (linear correlation) between X and Y,with a value between-1 and 1.A value of 1 indicates a perfectly positive correlation between the two random variables;a value of -1 indicates a perfectly negative correlation between the two random variables;a value of 0 indicates a linear correlation between the two random variables.

1.4.3 Stepwise linear regression

Stepwise regression analysis is a regression analysis method that selects independent variables to establish the optimal regression equation.The optimal regression equation refers to a regression equation with all independent variables that have a significant impact on the dependent variable,and without independent variables that have no significant impact on the dependent variable.The process involves gradually introducing independent variables while examining whether the independent variables that have been introduced into the model still have statistical significance,and to eliminate them if them do not.The general form of the multiple linear regression model is:Yi=β0+β1xi+β2xi+…+βkxi+μi,i=1,2,...,n,wherekis the number of explanatory variables,and is the regression coefficient.The parameter estimation of the multiple linear regression model,like that of the one-dimensional linear regression equation,also uses the least-squares method to solve the parameters while requiring the sum of squared errors (Σe) to be minimized.

1.4.4 Paired t-test

Assuming that the difference between the two sets of paired indicators to be tested isdi=x1i-The paired t-test is to test whether the difference between the means of the paired samples is zero,that is,whether the mean ofdiis zero,then the problem transforms into a singlesample t-test.Under the assumption of normality and equal variance,the mean value of the difference is:

For a two-sided test of whether the mean value of the difference is 0,if,then we reject the null hypothesis (H0∶μ1=μ2),indicating that the sample mean is significantly different from the population mean.Otherwise,we fail to reject the null hypothesis.

1.4.5 One-way ANOVA

One-way ANOVA is used to study whether different levels of a control variable have a significant impact on the observed variable.The total variation of the observed variable is decomposed into sum of squares for factor A and sum of squares for error(SST=SSA+SSE).By comparing the proportions of total variation and its component,we can infer whether the control variable has a significant impact on the observed variable.The F statistic is the test statistic used in ANOVA.Assuming that the samplesx1i,x2i…xnifrom each levelAi(i=1,2,…,…,k) is from normal distribution N (μi,σ2),whereμiand σ2are unknown,the F statistic is defined as:

n is the total number of observations and k is the number of levels of the control variable.At a given significance level α,if F>Fα,we reject the null hypothesis and conclude that the means of each group are significantly different.Otherwise,we fail to reject the null hypothesis and conclude that the means are not significantly different.

2 Results and Discussion

2.1 Subjects’ basic information

The 118 subjects,age distribution: 17 from 18 to 22 years old,19 from22 to 26 years old,23 from 26 to 30 years old,27 from 30 to 34 years old,14 from 34 to 38 years old,14 from 38 to 42 years old,and 4 from 42 to 46 years old;skin type distribution: 26 dry skins,18 combination/dry skins,23 normal skins,25 combination/oily skins,and 26 oily skins.The skin problems are shown in Table 1.

Table 1.Subjects’ skin problems

2.2 24 evaluators assess consistency

The Kendall coordination coefficient test was used to study the evaluation consistency,with 102 evaluations.From the results,the Kendall coordination coefficient test showed significance,p=0.000<0.05,which means that the evaluations of 24 evaluators are correlated,i.e.,it indicates that the evaluations are consistent.Also,the Kendall coordination coefficient was 0.678,between 0.6 and 0.8,which indicates a strong consistency.

2.3 Correlation between apparent age and skin characterization data

The results of the correlation between apparent age and their skin assessment results of 118 subjects are shown in Table 2.As shown in Table 2,there were significant positive correlations between apparent age and forehead lines,texture,UV spots,brown spots,undereye fine lines,nasolabial folds -expert clinical assessment,yellowness -expert clinical assessment,and chroma saturation -expert clinical assessment.There were significant negative correlations between apparent age and blackheads -expert clinical assessment and lightness.That is,higher apparent age was associated with more forehead lines,texture,UV spots,brown spots,fine lines,undereye fine lines,deeper nasolabial folds,more yellow skin,higher saturation,and fewer blackheads.

Table 2. Correlation analysis between apparent age and skin characterization parameters

2.3 Linear regression analysis of apparent age

The results of this study were further analyzed by multiple regression between each skin characterization index and apparent age and are shown in Table 3. The resulting model equation was: apparent age=64.042+0.702*texture+0.414*UV spots -0.121*porphyrins+0.221*undereye fine lines+0.084*average water content -1.066*acne marks -expert clinical assessment+0.977*nasolabial folds -expert clinical assessment -1.037*lightness+0.341*hue.

Table 3. Multifactorial linear regression analysis between apparent age and skin characterization indexes (n=115)

ItsR2value was 0.704,implying at texture,UV spots,porphyrins,undereye fine lines,average water content,acne marks -expert clinical assessment,nasolabial folds -expert clinical assessment,lightness and hue could explain 70.4% of the variation in apparent age.The model passed the F-test (F=27.740,p=0.000<0.05),indicating that the model is valid.In addition,the test for model multicollinearity found that all VIF values in the model were less than 5,implying that there was no co-collinearity problem;and the D-W value was around 2,indicating that the model did not have autocorrelation and there was no correlation between the sample data,the model fits well.

The final detailed analysis shows that the regression coefficient value of texture was 0.702(t=4.601,p=0.000<0.01),which means that texture has a significant positive effect on apparent age.

The regression coefficient value of UV spots was 0.414 (t=7.189,p=0.000<0.01),which means that UV spots have a significant positive effect on apparent age;the regression coefficient value for porphyrins was -0.121 (t=-3.736,p=0.000<0.01),implying a significant negative effect of porphyrins on apparent age;the regression coefficient value for undereye fine lines was 0.221 (t=5.620,p=0.000<0.01),implying a significant positive effect of undereye fine lines on apparent age;the regression coefficient value for average water content was 0.084 (t=2.486,p=0.015<0.05),implying a significant positive effect of average water content on apparent age;the regression coefficient value for acne marks -expert clinical assessment was -1.066 (t=-3.408,p=0.001<0.01),implying a significant negative effect of acne marks on apparent age;the regression coefficient value for nasolabial folds -expert clinical assessment was 0.977(t=2.670,p=0.009<0.01),implying that nasolabial folds have a significant positive effect on apparent age;the regression coefficient value for lightness was -1.037 (t=-4.479,p=0.000<0.01),implying a significant negative effect of lightness on apparent age;and the regression coefficient value for hue was 0.341 (t=3.116,p=0.002<0.01),implying a significant positive effect of hue on apparent age.

The scatter plot reveals that the model predicted values are in good agreement with the actual apparent age.The paired t-test was used to study the variability of the experimental data.As can be seen from Table 4,the paired data did not show a difference between the apparent age and the model predicted age (p>0.05),i.e.,the model predicted age results were not significantly different from the apparent age.

Figure 1. Scatter plot of apparent age vs. model-predicted age

Table 4.Paired test of apparent and predicted age

2.5 City-to-city comparison of apparent age and true age difference

The comparison of the difference between the true age and apparent age of subjects in different cities is shown in Table 5.The apparent age of subjects in Chengdu was closer to their true age,i.e.,they looked relatively younger,compared to other cities,but there was no statistical difference between cities.

Table 5.City-to-city comparison of true age and apparent age difference

2.6 Effect of lifestyle factors on apparent age

The effect of household income on apparent age is shown in Table 6.As seen in Table 6,subjects with a monthly income of 25,000RMB or more had relatively lower apparent age compared to those with a monthly household income of less than 25,000RMB,but there was no significant difference.

Table 6. The relationship between the difference between true and apparent age and household income

The effect of eating sweets on apparent age is shown in Table 7.From Table 7,it shows no significant correlation between eating sweets and apparent age.

Table 7. The relationship between the difference between true and apparent age and the frequency of eating sweets

The effect of cooking frequency on apparent age is shown in Table 8.From Table 8,it shows no significant correlation between cooking and apparent age.

Table 8.The relationship between the difference between true and apparent age and the frequency of cooking

The effect of drinking milk on apparent age is shown in Table 9.From Table 9,subjects who drank milk more frequently had a relatively lower apparent age,but it was not significant.

Table 9. The relationship between the difference between true and apparent age and the frequency of drinking milk

The effect of staying up late on apparent age is shown in Table 10.From Table 10,subjects with the habit of staying up late had a relatively higher apparent age,but there is no significance.

Table 10. The relationship between the difference between true and apparent age and the habit of staying up late

The effect of smoking on apparent age is shown in Table 11,From table 11,subjects who smoked had a relatively higher apparent age compared to those who did not,but there is no significance.

Table 11. The relationship between the difference between true and apparent age and the habit of smoking

The effect of using face cream on apparent age is shown in Table 12.From Table 12,subjects who used face cream had relatively lower apparent age,i.e.,subjects with the habit of using face cream looked younger compared to those without the habit.

Table 12.The relationship between the difference between true and apparent age and the habit of using face cream

3 Conclusion

By comparing the correlation between the apparent age of the subjects as judged by the assessors based on standard photos of the subjects and objective data of the subjects’ skin characterization parameters,as well as the lifestyle-related characteristics of the subjects’ collected by questionnaire,the following conclusions were obtained.

Forehead lines,texture,UV spots,brown spots,fine lines,undereye fine lines,laugh line -expert clinical assessment,yellowness -expert clinical assessment,and chroma saturation were significantly and positively correlated with apparent age.Blackheads -expert clinical assessment and lightness were significantly and negatively correlated with apparent age.That is,people with higher apparent age tend to have more forehead lines,texture,UV spots,brown spots,fine lines,undereye fine lines,deeper nasolabial folds and more yellow skin,higher skin tone saturation,lower skin tone lightness and fewer blackheads.

The apparent age regression model based on the subjects and test dimensions of this study was obtained by multifactorial regression analysis as:apparent age=64.042+0.702*texture+0.414*UV spots -0.121*porphyrin+0.221*undereye fine lines+0.084*water content-1.066*acne marks +0.977*nasolabial folds -1.037*lightness+0.341*hue.

That is,fairer,lower hue,fewer UV spots,texture,undereye fine lines,nasolabial folds,water content,more porphyrin and acne marks result in younger-looking skin.More acne marks make one look younger may be that acne marks and acne problems tend to occur during adolescence,thus creating a negative correlation between acne marks and apparent age.The above study is informative for the development of anti-aging cosmetic products,as well as the investigation into anti-aging efficacy.

The comparative study of lifestyle and the difference between apparent age and true age discovered that the habit of using face cream has a significant effect on the apparent age and true age difference,a finding that is consistent with Mayes et al.[6],suggesting that using face cream can,to some extent,help delay visual skin-aging and look younger.

Finally,this study found that the difference between apparent age and true age was somewhat different across the cities studied,but not statistically significant;subjects who smoked,stayed up late,did not drink milk,and had lower income were found to have a relatively higher apparent age compared to their counterpart group,but again not statistically significant.The reason for this may be due to the limitation of the sample size of this study,so we still need further studies in the future with larger sample sizes for the above comparison and impact studies.