Blockchain-Enabled Image Provenance Authentication for Camera-Origin vs. AI-Synthesized Media: Architecture and Validation

Abstract

With the blistering development of generative artificial intelligence and especially generative adversarial networks (GANs), as well as diffusion models, there is now an unprecedented difficulty in identifying fundamental limitations [2], [34], [53]. whether a particular image was taken by a camera or created by artificial intelligence. The paper outlines a holistic image authentication system based on blockchain that incorporates a multi-modal feature extraction method with blockchain-authenticated provenance verification that can be trusted to label camera images as original. Our four-layer architecture proposal is a combination of: (1) cryptographic image hashing with SHA-256, SHA-512, BLAKE2b and perceptual hashing; (2) a 50-dimensional forensic feature image consisting of color statistics, frequency domain (FFT) analysis, edge/texture descriptors, and image quality measures, (3) machine learning classification with ensembles of Random Forest, The experimental validation on a dataset of 196 images (96 real camera photographs by Unsplash and 100 AI-generated images with known GAN artifacts) shows that our ensemble model is able to detect the synthetic content with 97.96% accuracy and is resilient to typical post-processing manipulations, which provides a scalable and tamper-proof method to preserve the authenticity of digital images.