AI Skin Analysis · Privacy & GDPR · 15 October 2026
Di Elena Setaro
Choosing an AI skin analysis platform requires evaluating three key areas: AI skin analysis accuracy across different skin phototypes, facial image privacy and data protection requirements. For beauty brands and skincare retailers, these considerations help determine whether a technology is suitable for their customers and intended use.
AI skin analysis uses artificial intelligence and computer vision to evaluate visible skin characteristics from images such as smartphone selfies.
However, the ability to analyse blemishes, redness or wrinkles does not automatically demonstrate equivalent performance across all skin phototypes or establish how customer images are protected.
This guide focuses on the questions beauty brands should ask when evaluating AI skin analysis accuracy and privacy. For a broader overview of platform features, personalisation and integration options, explore Dermaself's guide to white-label AI skin analysis platforms.
Performance across different skin phototypes should be evaluated using validation results that identify which phototypes were tested and which skin-analysis functions were measured. Results obtained from one group should not automatically be interpreted as evidence of equivalent performance across every group.
Skin phototype and skin tone describe different characteristics.
Skin phototype
Il Fitzpatrick classification categorises skin according to its typical response to ultraviolet exposure.
Skin tone
Skin tone refers to the visible colour of the skin.
These concepts are related, but they are not interchangeable. For beauty brands evaluating AI skin analysis, the distinction matters because validation data referring to specific phototypes does not necessarily establish performance across all visible skin tones.
The phototypes represented in a validation sample determine which populations are directly covered by the reported results.
For example, a study conducted on phototypes II–IV does not automatically demonstrate equivalent performance on phototypes I, V or VI. This does not establish that the technology performs poorly on other phototypes. It means that the particular evaluation does not provide evidence for those groups.
Brands serving customers with diverse skin characteristics should therefore examine the population included in each provider's validation.
When evaluating a platform, brands should ask:
Which skin phototypes were included in testing?
How many images or participants were evaluated?
Were validation images excluded from model training?
Which skin characteristics were measured?
Are results available separately for different groups?
What limitations does the provider identify?
These questions make it easier to understand the scope of a reported performance result.
Dermaself 2026 internal validation
Dermaself evaluates the performance of its AI Skin Analysis technology through internal validation focused on specific skin-analysis capabilities, including redness assessment and wrinkle localisation. The validation process uses images not included in model training.
98%
di concordanza entro un grado nella valutazione del rossore
92%
di accuratezza nella localizzazione delle rughe
Rossore: assessments differed from the reference evaluation by no more than one grade in 98% of cases, based on 121 faces not used for model training, across skin phototypes II–IV.
Rughe: the model identified the correct facial area containing a wrinkle in 92% of cases, across more than 5,000 analysed wrinkles.
Source: Dermaself internal validation methodology, 2026. These findings describe the performance of specific AI capabilities, rather than representing an overall accuracy score or demonstrating equivalent performance across all skin phototypes.
For a broader explanation of Dermaself's technical validation, see the white-label AI skin analysis platform guide.
AI skin analysis accuracy should be evaluated by examining what was measured, how the evaluation was conducted and which data were used. A performance percentage is more informative when the underlying task and methodology are clearly explained.
Different AI skin analysis functions can require different forms of evaluation. For example, identifying the facial area containing a wrinkle is a different task from assessing the intensity of redness. As a result, percentages describing these functions should not be treated as directly interchangeable.
A useful validation process evaluates performance using images that were not included in the model's training data. Dermaself applies this approach in its internal redness and wrinkle evaluations. When comparing platforms, brands should ask whether the reported results were obtained using data excluded from training.
The meaning of a performance result depends on the evaluation criteria. For example, Dermaself's redness result refers to assessments differing from the reference evaluation by no more than one grade. It does not mean that 98% of all redness assessments were exact matches.
Similarly, the wrinkle-location result measures the identification of the correct facial area, rather than every possible aspect of wrinkle analysis. Clearly defined metrics help brands understand what a platform's validation results demonstrate.
Lighting, exposure, white balance and smartphone camera processing can affect how skin appears in a photograph. These variations are relevant to image-based analysis because the AI evaluates visual information captured by the camera.
Image normalisation is designed to reduce differences caused by image-capture conditions and create more consistent inputs for analysis.
Research collaboration
Dermaself has worked with Alta Scuola Politecnica, the joint honours programme of Politecnico di Milano and Politecnico di Torino, on research addressing selfie image normalisation. The project explored an additional image-processing pipeline combining AI-driven illumination correction with classical image-processing algorithms.
The research is described in Cosmetics Business: Advancing Privacy-Preserving AI Skin Analysis – Dermaself and Alta Scuola Politecnica.
The project was developed to complement Dermaself's existing AI skin analysis technology. Its description should not be interpreted as confirmation that every research component is available in every commercial implementation.
Facial image privacy requires attention to both the visible skin information needed for analysis and the facial characteristics that may identify an individual. Technologies designed to reduce identifiable information should be evaluated according to their actual processing methods and documented capabilities.
A smartphone selfie can contain useful information about skin characteristics alongside features associated with the person's identity. This creates a challenge for image-based technologies: retaining the visual details needed for skin analysis while reducing identifying facial information.
Traditional image-processing approaches, including masking and blurring, can remove or obscure visible facial features. However, these techniques may also remove skin details relevant to analysis. This limitation was one of the questions explored in the Dermaself and Alta Scuola Politecnica research project.
The Alta Scuola Politecnica project developed a research approach based on 3D facial reconstruction. The technology was designed to separate identity-related facial information from skin characteristics useful for analysis. Rather than relying exclusively on masking or blurring, the research explored structural transformations of facial representations while preserving visible skin details such as lesions, redness, pigmentation and texture.
The project is described in Cosmetics Business's article on Dermaself and Alta Scuola Politecnica.
This work represents research intended to expand Dermaself's technological capabilities. It should not be interpreted as confirmation of irreversible anonymisation or as evidence that the technology is already included in every Dermaself deployment.
Before implementing a selfie-based platform, brands should ask:
What facial information is collected?
Are original images retained after processing?
Which techniques are used to reduce identifiable information?
Can processed images still be linked to an individual?
Which privacy technologies are available in the proposed implementation?
What documentation supports the provider's privacy claims?
These questions help brands distinguish research objectives, technical capabilities and actual commercial deployment.
GDPR evaluation should cover the complete processing of personal data, not only the technology used to modify facial images. A privacy-focused image-processing method does not, by itself, establish that an AI skin analysis platform complies with all applicable data protection requirements.
Il General Data Protection Regulation (EU) 2016/679 provides the European legal framework for the protection of personal data. For beauty brands considering selfie-based AI skin analysis, the following questions provide a practical starting point for reviewing a provider's documentation.
What personal data is collected?
Brands should establish which information the platform processes and for which purposes, including how facial images and any associated information are used within the proposed customer experience.
How are images stored and retained?
Brands should ask whether original selfies and processed images are stored, where they are stored and for how long, based on the provider's actual documentation rather than assumptions about the technology.
Who can access the information?
Brands should clarify which organisations and authorised personnel can access the data. The relevant security arrangements should be documented for the intended implementation.
How are customer data protection rights managed?
The provider and brand should clarify how applicable requests concerning personal data are handled, including the responsibilities of the organisations involved.
Which organisation is responsible for each processing activity?
The parties should establish their respective data protection roles and responsibilities. These arrangements depend on how the service is implemented and how the data is processed.
Data protection should be considered during the design and implementation of systems that process personal data.
The European Data Protection Board provides Guidelines 4/2019 on Article 25, Data Protection by Design and by Default. For beauty brands, the practical implication is that privacy requirements should be evaluated before introducing selfie-based analysis into the customer journey.
This guide does not assess Dermaself's complete GDPR compliance status, retention arrangements or contractual data-processing responsibilities for any specific implementation. These details should be verified using the relevant official documentation.
Beauty brands should not assume that every facial image is processed for biometric identification. They should establish what technical processing the platform performs and whether identifying an individual is among its purposes.
Il GDPR addresses biometric data in Article 4 and includes specific provisions concerning biometric data processed for the purpose of uniquely identifying a natural person in Article 9.
For a platform using selfies, relevant questions include:
Is the image used to analyse visible skin characteristics?
Does the system also perform identity identification or verification?
What information is derived from the image?
Are results associated with identifiable customer information?
What legal assessment supports the proposed processing?
Dermaself describes its technology as analysing visible skin characteristics to support personalised skincare recommendations. However, the legal classification of each processing activity depends on the specific customer deployment.
For that reason, this guide does not draw conclusions about Dermaself's biometric data processing or GDPR compliance.
Beauty brands should request evidence about technical validation, skin phototype coverage and facial image processing before selecting an AI skin analysis provider. The following checklist brings together the questions covered in this guide.
Area di valutazione
Question for the provider
Skin phototypes
Which phototypes were included in validation?
Skin tone coverage
Which visible skin tones were represented?
Validation sample
How many images or participants were evaluated?
Performance metrics
Which function was measured, and how?
Training separation
Were validation images excluded from model training?
Image quality
How are lighting and camera variations addressed?
Facial privacy
Which identifiable characteristics are retained or reduced?
Research versus deployment
Which technologies are available in the proposed solution?
Data storage
Are original or processed selfies stored, and for how long?
GDPR documentation
What documentation explains processing purposes and responsibilities?
Customer rights
How are applicable data protection requests handled?
These questions help digital, e-commerce and innovation teams assess the evidence behind a provider's claims.
For broader criteria such as API integration, white-label deployment, ingredient-based recommendations and product catalogue compatibility, read Dermaself's guide to evaluating an AI skin analysis platform.
Dermaself connects selfie-based analysis of visible skin characteristics with ingredient information and products from beauty brands' and retailers' catalogues to support personalised skincare recommendations.
Dermaself documents internal validation results for specific skin-analysis functions and has carried out research collaborations addressing image consistency and privacy-focused facial processing.
The collaboration with Alta Scuola Politecnica explored two complementary technologies: selfie image normalisation e 3D-based facial anonymisation.
For further information, explore Dermaself's technology and the Cosmetics Business article on the academic research collaboration.
Brands considering a specific implementation should request the corresponding technical and data protection documentation.
Learn how Dermaself connects AI skin analysis with cosmetic ingredient information and personalised skincare recommendations for beauty brands and retailers.
Risposte rapide alle domande più comuni
Performance across all skin phototypes should not be assumed from results obtained on a limited group. Brands should examine which phototypes were included in validation and what the reported results measure.
Skin phototype describes the skin's typical response to ultraviolet exposure, while skin tone refers to its visible colour. They are related but distinct characteristics.
Dermaself reports internal validation results for specific functions, including redness assessment on 121 faces across phototypes II–IV and wrinkle-location evaluation across more than 5,000 analysed wrinkles.
Lighting, exposure and camera processing can change how skin appears in a selfie. Image normalisation research aims to reduce these differences and create more consistent inputs for AI analysis.
The research collaboration explored selfie image normalisation and a 3D-based facial anonymisation approach designed to preserve relevant skin information while reducing identity-related facial characteristics.
No. The complete processing activity and the relevant legal requirements must be assessed separately.
This article does not assess the complete GDPR compliance arrangements for any specific Dermaself implementation. Brands should request the applicable official documentation.
No. Dermaself describes its AI skin analysis as a technology supporting personalised skincare experiences and product recommendations, not as a medical diagnostic tool.
Integra l’analisi della pelle da selfie e le raccomandazioni skincare personalizzate nel tuo e-commerce o punto vendita, collegando le esigenze della pelle ai prodotti disponibili nel tuo catalogo.
Prenota una Demo