Treffer: Region-aware digital makeup synthesis via semantic face parsing and multi-region colour distribution transfer

This project presents a region-aware digital makeup framework that integrates deep-learning–based facial segmentation with multi-region colour distribution transfer to achieve photorealistic and customisable virtual cosmetic effects. The system employs a pre-trained BiSeNet model to generate high-resolution semantic face masks, enabling precise localisation of key facial regions such as skin, lips, eyes, and hair. Building upon these masks, the framework performs reference-guided colour adaptation using a probability density–based Whitening–Colouring Transform (WCT) in the CIE-Lab colour space, aligning regional chromatic statistics between a source portrait and a curated style exemplar. Additional modules, including anatomically guided eyeshadow placement and a perceptually coherent blush synthesis pipeline, further enhance realism while preserving natural texture and illumination. The prototype is deployed as a React–FastAPI application, supporting both one-click preset styles and fine-grained manual adjustments through layered cosmetics such as lipstick, blush, and eyeshadow. Evaluation across diverse facial profiles demonstrates strong generalisation, stable segmentation boundaries, and consistent photometric behaviour, though performance remains sensitive to pose variation and extreme lighting conditions. Overall, the results show that combining semantic parsing with statistical colour modelling provides an interpretable, efficient, and flexible approach for high-fidelity digital makeup generation, offering a practical foundation for future AR, beauty enhancement, and virtual try-on applications. ; Bachelor's degree