홍채 색상 변경 서비스를 위한 경량화된 상호작용 시스템

초록

With the recent increase in demand for augmented reality-based beauty technologies and personalized user experiences, the needfor automated and real-time iris color changing or virtual color contact lens try-on services has grown significantly. However, conventional0software-based iris color changing methods are limited in terms of real-time application and user convenience due to the boundaryprecision and latency of automation tools, the proficiency required for manual editing, and the reliance on repetitive tasks. This studyproposes an automated system that integrates face, eye, and iris detection as well as masking and color modification within a singlepipeline, based on a lightweight deep learning model utilizing a multi-task loss function. Experimental evaluations show excellentperformance, with median blind reading scores of 4.62/5 for masking accuracy and 4.51/5 for color naturalness, and a 98% match ratefor iris count. In terms of efficiency, the median per-frame latency was measured at 142.00 ms (7.04 FPS), and system memory usagewas stably maintained within a maximum of 95.852 MiB.

키워드