Forensic Analysis of AI-Generated Image Alterations Using Metadata Evaluation, ELA, and Noise Pattern Analysis

This study develops a forensic workflow to assess the authenticity of digital images, addressing the challenge of distinguishing AI-generated content from real photographs. The goal is to analyze metadata, compression behavior, and noise characteristics to identify synthetic images. The dataset includes eight images: two original Xiaomi 14T Pro photos and six AI-generated variants from Gemini, ChatGPT, and Copilot. Metadata was extracted using ExifTool version 13.25 on Kali Linux, while Error Level Analysis (ELA) and Noise Pattern Analysis (NPA) were performed with consistent parameters on the Forensically platform. Authentic images displayed complete EXIF metadata, uniform compression patterns, and stochastic sensor noise. In contrast, AI-generated images lacked EXIF data, included XMP or C2PA provenance, exhibited localized compression anomalies, and showed smoother, more structured noise patterns. The study presents a practical and reproducible forensic workflow that integrates metadata evaluation, ELA, and noise analysis to detect synthetic content. The findings demonstrate that despite their visual realism, AI-generated images still leave detectable forensic traces, offering valuable tools for image authenticity verification.