Byte-level generative predictions for forensics multimedia carving

2026-04-13

AI Summary

This study addresses the challenge of reconstructing fragmented multimedia files in digital forensics when filesystem metadata is unavailable. To overcome this limitation, the authors propose a generative model–based file carving approach that introduces, for the first time, a byte-level generative Transformer (bGPT) to multimedia file reconstruction. Rather than relying solely on classification strategies, the method predicts subsequent bytes of BMP image fragments to synthesize missing data. The quality of the generated content is rigorously evaluated using multiple metrics, including cosine similarity, structural similarity index (SSIM), chi-square distance, and Jensen–Shannon divergence. Experimental results demonstrate that the proposed technique effectively captures byte-level patterns and significantly improves the accuracy of fragment matching within unallocated disk space.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Computer Vision: Multi-modal VisionData Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data

Application Category

Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAGWeb Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web dataGraph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Digital forensic investigations often face significant challenges when recovering fragmented multimedia files that lack file system metadata. While traditional file carving relies on signatures and discriminative deep learning models for fragment classification, these methods cannot reconstruct or predict missing data. We propose a generative approach to multimedia carving using bGPT, a byte-level transformer designed for next-byte prediction. By feeding partial BMP image data into the model, we simulate the generation of likely fragment continuations. We evaluate the fidelity of these predictions using different metrics, namely, cosine similarity, structural similarity index (SSIM), chi-square distance, and Jensen-Shannon divergence (JSD). Our findings demonstrate that generative models can effectively predict byte-level patterns to support fragment matching in unallocated disk space.

Problem

• file carving
• fragmented multimedia files
• missing data prediction
• digital forensics
• byte-level prediction

Innovation

• generative modeling
• byte-level prediction
• multimedia carving
• digital forensics
• transformer