Implementation and Performance Analysis of an Identity-Aware Deepfake Forensic Framework Using 128-Dimensional Biometric Embeddings and CNN-Based Verification

Existing deepfake detection systems lack identity-awareness and cannot determine the potential victim associated with manipulated media. This paper presents an identity-aware forensic framework that integrates 128-dimensional facial embedding retrieval using vector similarity search with CNN-based verification. Following a Retrieve-Verify architecture, the system uses FAISS indexing for biometric retrieval followed by CNN authenticity classification. The system achieved a validated accuracy of 83.6% on a dataset of 5,000 images, transforming detection into a forensic attribution problem.