AI TraceFinder — Forensic Scanner Identification

Overview

AI TraceFinder is a forensic machine learning platform that identifies the source scanner device used to digitize a document or image. Each scanner (brand/model) introduces unique noise, texture, and compression artifacts that serve as a fingerprint. By analyzing these patterns, AI TraceFinder enables fraud detection, authentication, and forensic validation in scanned documents.

---

Goals & Objectives

• Collect and label scanned document datasets from multiple scanners
• Robust preprocessing (resize, grayscale, normalize, denoise)
• Extract scanner-specific features (noise, FFT, PRNU, texture descriptors)
• Train classification models (ML + CNN)
• Apply explainability tools (Grad-CAM, SHAP)
• Deploy an interactive app for scanner source identification
• Deliver accurate, interpretable results for forensic and legal use cases

---

Methodology

1. Data Collection & Labeling

   • Gather scans from 3–5 scanner models/brands
   • Create a structured, labeled dataset

2. Preprocessing

   • Resize, grayscale, normalize
   • Optional: denoise to highlight artifacts

3. Feature Extraction

   • PRNU patterns, FFT, texture descriptors (LBP, edge features)

4. Model Training

   • Baseline ML: SVM, Random Forest, Logistic Regression
   • Deep Learning: CNN with augmentation

5. Evaluation & Explainability

   • Metrics: Accuracy, F1-score, Confusion Matrix
   • Interpretability: Grad-CAM, SHAP feature maps

6. Deployment

   • Streamlit app → upload scanned image → predict scanner model
   • Display confidence score and key feature regions

---

Actionable Insights for Forensics

• Source Attribution: Identify which scanner created a scanned copy of a document.
• Fraud Detection: Detect forgeries where unauthorized scanners were used.
• Legal Verification: Validate whether scanned evidence originated from approved devices.
• Tamper Resistance: Differentiate between authentic vs. tampered scans.
• Explainability: Provide visual evidence of how classification was made.

---

Architecture (Conceptual)

Input ➜ Preprocessing ➜ Feature Extraction + Modeling ➜ Evaluation & Explainability ➜ Prediction App

---

⏳ 8-Week Roadmap (Milestones)

• W1: Dataset collection (min. 3–5 scanners), labeling, metadata analysis
• W2: Preprocessing pipeline (resize, grayscale, normalize, optional denoise)
• W3: Feature extraction (noise maps, FFT, LBP, texture descriptors)
• W4: Baseline ML models (SVM, RF, Logistic Regression) + evaluation
• W5: CNN model training with augmentation, hyperparameter tuning
• W6: Model evaluation (accuracy, F1, confusion matrix) + Grad-CAM/SHAP analysis
• W7: Streamlit app development → image upload, prediction, confidence output
• W8: Final documentation, results, presentation, and demo handover

---

Suggested Project Structure

ai-tracefinder/
├─ app.py              
├─ src/
│  ├─ ingest/           
│  ├─ preprocess/        
│  ├─ features/          
│  ├─ models/            
│  ├─ explain/           
│  └─ utils/             
├─ data/                 
├─ notebooks/            
├─ reports/              
└─ README.md