Caricature Image Generation from Near-Infrared Facial Images

Facial recognition systems, widely used as a contactless authentication method in modern society, have some major limitations, including the dependency of visible light cameras on external environmental conditions and the risk of personal information theft and abuse when displayed on screens. This paper introduces a novel approach to address these challenges. The proposed method utilizes near-infrared images as input which are robust to lighting variations and displays corresponding caricature images to achieve de-identification. Additionally, we analyze the role, limitation, and improvement of each sub-model within the proposed method and show that the proposed method significantly enhances the usability and security of conventional facial recognition systems, by comparing the translation quality and inference time across various versions of the implemented model.