Abstract

Facial recognition technology has become an important component of modern computer vision systems and is widely used for authentication, surveillance, access control, social media applications, and identity verification. However, the increasing use of automated facial recognition creates significant privacy risks when facial images are collected, processed, or shared without adequate user consent. Existing facial privacy protection approaches include adversarial perturbations, image transformations, semantic manipulation, personalised masks, adversarial makeup, and diffusion-based techniques. Although these methods can reduce the effectiveness of facial recognition, they may suffer from limitations related to image quality, transformation robustness, recognition model dependence, computational cost, and generalisation. This paper presents a conceptual adaptive hybrid framework for protecting facial images against unauthorised face recognition. The proposed framework combines semantic perturbation, adaptive pixel-level noise, and adaptive image filtering. The approach first analyses the facial image and its important features, then applies semantic-level modifications followed by content-aware pixel perturbation and image filtering. The objective is to reduce the recognizability of the protected image while maintaining acceptable visual quality. The paper reviews existing facial privacy protection approaches, identifies their limitations, and discusses an integrated framework that can adapt protection according to image characteristics and recognition threats. The proposed framework is intended as a conceptual research direction for privacy-preserving computer vision and does not claim a completed implementation or experimental accuracy result.

Keywords

How to Cite This Article

Rights