A Study on the Development and Application of Image-based Facial Skin Aging Diagnosis Technology Using A.I. Model
223
Presented by: Sae-ra Park
A Study on the Development and Application of Image-based Facial Skin Aging Diagnosis Technology Using A.I. Model
Sae-ra, Park1; Hyeokgon, Park1; Joongwon, Hwang2; Sangran, Lee2; Eunjoo, Kim1*;
1 Clinical Research Lab, AMOREPACIFIC R&I Center, Gyeonggi-do, Republic of Korea;
2 AI Solution Team, AMOREPACIFIC Corporation, Seoul, Republic of Korea
*eon827@amorepacific.com(E-mail address of the corresponding author)
Abstract
Introduction
Image analysis using artificial intelligence (A.I.) technology is being used in various fields, and the demand is increasing with the development of skin research, cosmetics industry, and personalization services. In particular, methods of using A.I. model that learned facial image for quantitative evaluation of facial skin aging have been attempted, but the correlation with human skin aging recognition and the suitability for skin aging research have not been confirmed. In this study, we developed an image-based facial aging diagnosis system to evaluate the overall facial skin aging with intuitive and quantified age levels using deep learning technology, and confirmed the correlation with the expert's judgment. In addition, we investigated the applicability of this system in skin aging change studies and efficacy evaluation of anti-aging cosmetics.
Methods
To develop a system for facial skin aging diagnosis, 11,000 facial images were collected from Korean volunteers aged 19 to 79 years. The A.I. facial skin aging diagnostic system was developed based on the SSR-Net(Soft Stagewise Regression Network) model which wass initialized with ImageNet Pretrained Weight and was developed to predict perceive age according to facial aging. 2,000 facial images were used for development tests to verify the diagnostic accuracy and performance of this system. In addition, 160 facial images were used by five clinical experts to determine the correlation between human aging awareness and the results of this system. In addition, to confirm the applicability of this system to skin aging study and cosmetic efficacy evaluation, the facial images obtained from the long-term skin aging study and the efficacy evaluation study of anti-aging cosmetics were analyzed and compared with the results of the existing method.
Results
As a result of confirming the skin aging diagnosis performance of the system in facial images of various ages, the system’s predicted age showed a difference of biological age and ±3 years old, and it showed a significant correlation with the actual facial age determined by clinical experts. In addition, it was possible to quantify the aging degree of the participants by analyzing the aging of the entire face in the 4-year long skin aging study. The group using retinol-containing products, which are anti-aging components, showed lower predicted facial age than the group without use, and the distribution of facial aging changes in the group was also low. And as a result of evaluating images of participants who used anti-aging products with syringaresinol, hydrolyzed ginseng saponins, and bioflavonoids as the main active ingredients, it was showed that the predicted facial age was significantly lowered compared to before use.
Discussion and Conclusion
The facial skin aging diagnosis system developed by deep learning can evaluate overall facial skin aging changes using only an optical facial image and predict facial skin age based on this result. It is also possible to be utilized to long-term skin change studies or to evaluate the efficacy of anti-aging cosmetics. Our facial skin aging diagnosis system was verified by comparison with expert evaluation and it showed high accuracy. Compared to the conventional skin evaluation equipment or the visual assessment of experts, it has the advantage of being able to evaluate the entire face more quickly, objectively and at a low cost. Therefore, we expect that the image-based facial aging diagnosis system using A.I. can be used not only for skin change study or cosmetic evaluation research, but also for personalization services such as customized cosmetics.
Keywords:
Artificial intelligence(A.I.); facial aging; skin; deep learning; diagnosis
Sae-ra, Park1; Hyeokgon, Park1; Joongwon, Hwang2; Sangran, Lee2; Eunjoo, Kim1*;
1 Clinical Research Lab, AMOREPACIFIC R&I Center, Gyeonggi-do, Republic of Korea;
2 AI Solution Team, AMOREPACIFIC Corporation, Seoul, Republic of Korea
*eon827@amorepacific.com(E-mail address of the corresponding author)
Abstract
Introduction
Image analysis using artificial intelligence (A.I.) technology is being used in various fields, and the demand is increasing with the development of skin research, cosmetics industry, and personalization services. In particular, methods of using A.I. model that learned facial image for quantitative evaluation of facial skin aging have been attempted, but the correlation with human skin aging recognition and the suitability for skin aging research have not been confirmed. In this study, we developed an image-based facial aging diagnosis system to evaluate the overall facial skin aging with intuitive and quantified age levels using deep learning technology, and confirmed the correlation with the expert's judgment. In addition, we investigated the applicability of this system in skin aging change studies and efficacy evaluation of anti-aging cosmetics.
Methods
To develop a system for facial skin aging diagnosis, 11,000 facial images were collected from Korean volunteers aged 19 to 79 years. The A.I. facial skin aging diagnostic system was developed based on the SSR-Net(Soft Stagewise Regression Network) model which wass initialized with ImageNet Pretrained Weight and was developed to predict perceive age according to facial aging. 2,000 facial images were used for development tests to verify the diagnostic accuracy and performance of this system. In addition, 160 facial images were used by five clinical experts to determine the correlation between human aging awareness and the results of this system. In addition, to confirm the applicability of this system to skin aging study and cosmetic efficacy evaluation, the facial images obtained from the long-term skin aging study and the efficacy evaluation study of anti-aging cosmetics were analyzed and compared with the results of the existing method.
Results
As a result of confirming the skin aging diagnosis performance of the system in facial images of various ages, the system’s predicted age showed a difference of biological age and ±3 years old, and it showed a significant correlation with the actual facial age determined by clinical experts. In addition, it was possible to quantify the aging degree of the participants by analyzing the aging of the entire face in the 4-year long skin aging study. The group using retinol-containing products, which are anti-aging components, showed lower predicted facial age than the group without use, and the distribution of facial aging changes in the group was also low. And as a result of evaluating images of participants who used anti-aging products with syringaresinol, hydrolyzed ginseng saponins, and bioflavonoids as the main active ingredients, it was showed that the predicted facial age was significantly lowered compared to before use.
Discussion and Conclusion
The facial skin aging diagnosis system developed by deep learning can evaluate overall facial skin aging changes using only an optical facial image and predict facial skin age based on this result. It is also possible to be utilized to long-term skin change studies or to evaluate the efficacy of anti-aging cosmetics. Our facial skin aging diagnosis system was verified by comparison with expert evaluation and it showed high accuracy. Compared to the conventional skin evaluation equipment or the visual assessment of experts, it has the advantage of being able to evaluate the entire face more quickly, objectively and at a low cost. Therefore, we expect that the image-based facial aging diagnosis system using A.I. can be used not only for skin change study or cosmetic evaluation research, but also for personalization services such as customized cosmetics.
Keywords:
Artificial intelligence(A.I.); facial aging; skin; deep learning; diagnosis