Explore the physics of facial skin imaging
This educational research app extracts relative optical descriptors from a facial image and explores a two-phase TETRA-inspired homogenization model. It does not measure skin permittivity, hydration, sebum, disease, or clinical skin type.
Privacy: photographs and analysis are processed in your browser, without uploads to a server by this app. No accounts, tracking scripts, or automatic data transmission are included. You control optional JSON export. Avoid including other people in the frame.
1 · Capture facial image
Use even frontal lighting, no beauty filters, and keep the whole face visible. On a phone, camera access requires HTTPS or localhost.
Position your face inside the oval. Avoid makeup filters, direct sunlight and strong shadows. Remove other people from the frame.
No image captured.
Region sampling is approximate and manually adjustable, not automated face segmentation. Adjust the sampling frame to match your face before analyzing.
Adjust facial sampling frame
Move and resize the estimated facial region so that forehead and cheek sampling squares lie on skin, not hair, eyes or background. Colored boxes are sampling regions only.
2 · Optional cosmetic questionnaire
3 · Results
Image-derived optical descriptors
TETRA homogenization: explicit conclusion
Maxwell–Garnett illustrative two-phase model: εhost = 2.25 and εinclusion = 3.24 are assumed demonstration constants, not inferred skin properties. The camera features do not uniquely determine dielectric parameters. The displayed Fresnel reflectance is a simulated normal-incidence interface response, not measured reflectance.
How strongly does the theoretical model respond?
This sensitivity experiment changes only the assumed inclusion fraction. It is not fitted to your face and cannot identify skin type.
Provisional cosmetic skin-type profile
This separate result is based on questionnaire responses only. The TETRA simulation is deliberately not used to assign a skin-type label without calibration and reference-labeled validation data.
4 · Repeatability and reference-label study
After analyzing a photo, save its image-derived descriptors as a local study record. Retake another image in the same lighting and compare. No images are stored in this study table. Records remain in browser memory until you clear them or close the page.
No study records saved.
Repeatability uses only captures passing the preliminary image-quality gate. A standard deviation or coefficient of variation measures variation between images, not accuracy. Do not treat these as skin tissue properties.
Reference-label validation (exploratory)
Compare questionnaire-only predictions against independently assigned reference labels. This is an in-session descriptive check, not a validated accuracy study. Repeated images from one person are not independent subjects. The TETRA simulation is not a classifier.
5 · Calibration dataset readiness audit
Import the CSV exported by the study module. This tool checks sample coverage, quality and independent session counts. It does not train a classifier, estimate permittivity, or claim accuracy.
A session code must represent a distinct participant for the independent-participant counts to be meaningful. Do not put names or identifiable details in session codes.
Scientific and public-use information
The research objective is to investigate links between image-derived optical heterogeneity and physically meaningful effective-medium parameters. Future validation would require controlled reference measurements, repeatability testing, device and lighting comparisons, skin-tone diversity, and independently labeled skin types.
Inspired by the research themes of COST Action CA23125 TETRA. This exploratory application is not presented as an officially validated COST outcome or endorsed cosmetic product.
6 · TETRA inverse-model laboratory (synthetic data)
Simulation only. This module uses a user-selected hypothetical reflectance target. It does not derive reflectance from your smartphone photograph, and cannot infer skin permittivity or skin type.
This single-wavelength, normal-incidence model illustrates why inverse estimates depend on constituent assumptions. Real skin is layered, absorbing and scattering; RGB pixel values are not calibrated reflectance. A validated camera response model and independent reference measurements are required before fitting human skin.
7 · Camera-to-reflectance calibration bridge (experimental)
A smartphone RGB value is not physical reflectance. To explore the required calibration step, enter paired measured image luminance and independently measured reflectance (%) from calibration standards under fixed lighting and camera settings. This fits a simple empirical line, not a validated skin optical model.
An estimate is disabled when the latest image fails the quality gate or falls outside the calibration luminance range. This preliminary mapping ignores spectral response, geometry, skin scattering and camera auto-exposure; it must not be interpreted as measured skin reflectance or fed into a skin-type claim.