Region-specific calibration raises smartphone skin color accuracy to excellent

A new study from Sungwoo Kang and Jong-Kook Kim, published on arXiv (2512.21988), tackles the critical challenge of inter-device color reliability in smartphone-based dermatology. The team analyzed matched facial images from 965 Korean subjects captured by three devices: a professional DSLR camera, a consumer tablet, and a consumer smartphone. They benchmarked two calibration methods against the DSLR reference: a standard global linear Color Correction Matrix (CCM) and a region-specific CCM trained per anatomical region (e.g., forehead, cheek, chin), both applied in CIELAB color space.

The results show that standard global CCM already reduces inter-device color differences by 61-74%, achieving good reliability with an intraclass correlation coefficient (ICC) of 0.80 for Melanin Index (MI) and 0.78 for Individual Typology Angle (ITA). However, the region-specific CCM significantly outperformed it, raising MI ICC to 0.95 and ITA ICC to 0.93—both in the excellent reliability band. Importantly, the anatomical region itself was the largest pre-calibration variance contributor (η² = 0.18), surpassing source device variance (η² = 0.12), indicating that differences in skin tone across facial areas matter more than the device used. This finding suggests that smartphone dermatology can achieve clinically useful inter-device reliability with standard calibration, but region-aware calibration offers the largest remaining improvement, paving the way for more accurate remote skin assessments.