A User-Centric Context-Aware Framework for Real-Time Optimisation of Multimedia Data Privacy Protection, and Information Retention Within Multimodal AI Systems

This framework proposes a user-centric, context-aware, and ontology-driven privacy protection framework that automatically updates privacy decisions according to the relative privacy sensitivity of recognised entities, identified contexts, user-defined preferences, and the risk of exposure of Personally Identifiable Information (PII). The framework integrates state-of-the-art models, including YOLOv5, MTCNN, AlexNet, SlowFast R50, and CAER, for recognising faces, objects, scenes, actions and emotions in real time. Using a privacy ontology based on Contextual Integrity theory, it classifies entities into private, semi-private or public categories and enforces adaptive privacy levels through obfuscation techniques. A multi-level privacy model enables users to select between No Privacy, Low, Medium, High or Auto Privacy settings, alongside defining personal "red lines" (e.g., "always hide logos") that are strictly enforced regardless of context. The framework also proposes a Re-Identifiability Index (RII) using soft biometric features such as gait, hairstyle, clothing, skin tone, age and gender, to mitigate identity leakage and to support fallback protection when face recognition fails. Results of validation trials with 200 randomly selected users showed that user privacy remained protected effectively, with 85.2% of respondents finding the obfuscation operations highly effective and the other 14.8% stating that obfuscation was adequately effective. Amongst these, 71.4% considered the balance between privacy protection and usability very satisfactory and 28% found it satisfactory. GPU acceleration was deployed to enable real-time performance of these models by reducing frame processing time from 1200ms (CPU) to 198ms. This context-aware, ontology-driven framework offers a scalable, GDPR-compliant privacy protection framework that balances privacy protection with visual intelligibility and is suitable for a wide range of real-world applications including smart homes, healthcare, surveillance, and social media.