Privacy-preserving techniques in visual computing: applications, challenges, and future directions

Abstract

The proliferation of imaging technologies across healthcare, agriculture, environments, and consumer applications has heightened concerns regarding the privacy and security of visual data, which may contain biometric, medical, spatial, and location-specific information. Although federated learning (FL), differential privacy (DP), and secure multi-party computation (SMPC) have been reviewed as individual privacy-preserving approaches, existing surveys have predominantly examined these mechanisms in isolation from privacy-aware design principles. Moreover, limited attention has been given to imaging as a data modality, despite modality-specific privacy, and information leakage risks that differ from those associated with tabular and textual data. To address this research gap, this mini-review adopts a PRISMA-guided systematic review methodology to examine the literature on privacy-preserving approaches for imaging applications. An initial corpus of more than 5,000 records was screened and refined to a final set of 50 high-quality journal and conference publications based on publisher reliability, venue quality, and reporting rigor. The review establishes four dimensions of novelty in comparison with existing surveys and provides a characterization of five imaging sectors. For each sector, representative applications derived from implementations are examined alongside mathematically formalized challenges associated with federated learning, differential privacy, and secure multi-party computation. Trends in publication activity, application domains, and privacy-related challenges are analyzed through narrative and tabular presentations. Based on the synthesized evidence, the review contextualizes current adoption patterns and delineates prospective research directions for privacy-aware visual computing systems, distinguishing direct empirical observations from conceptual research trajectories.