Client-side image forensics

Is that document real? Find out in seconds.

DocuForge runs five forensic engines directly in your browser to expose edited, spliced and cloned regions in photos and documents. Nothing is uploaded — your files never leave your device.

Drop an image here, or click to chooseJPEG or PNG · analyzed locally in your browser

Five forensic engines, zero uploads

Error Level Analysis

Re-compresses the image and maps residuals — regions saved at a different quality glow.

Copy–Move

Block feature hashing with offset clustering catches regions cloned within the same image.

Noise Consistency

Laplacian residual mapping exposes spliced patches whose sensor noise doesn't match.

Metadata Forensics

Reads EXIF to flag editing software, timestamp conflicts and stripped camera data.

JPEG Compression

Estimates last-save quality from quantization tables to weigh the other signals.

1

Upload

Drop any photo, scan, receipt, certificate or ID-style document. It stays on your device.

2

Analyze

The engines run sequentially in-browser and paint tamper heatmaps over your image.

3

Report

Get a composite tamper-risk verdict and download a structured JSON evidence report.

Research layer

This repo also ships a PyTorch training + evaluation pipeline (CASIA v2, EfficientNet-B0 on ELA maps, AUC/F1/ROC harness) documented in training/ and docs/ — the deep-learning roadmap behind these classical engines.

Read the docs ↗