EAS 510 Assignment 1: Digital Forensics Apprentice

Dataset for building a rule-based expert system that matches modified images back to their originals.

Dataset Structure

EAS510_Assignment1/
├── originals/        # 10 original JPEG images
├── modified_images/  # 60 "easy" cases (single transformations)
├── hard/             # 60 "hard" cases (combined transformations)
└── random/           # 15 unrelated images (should be rejected)

Transformations

Easy Cases (modified_images/)

Each original has 6 modifications:

• Brightness enhancement
• JPEG compression
• 25% crop (center)
• 50% crop (center)
• 75% crop (center)
• PNG format conversion

Hard Cases (hard/)

Each original has 6 challenging modifications:

• v1: Off-center crop + compression
• v2: Crop + brightness + compression
• v3: Resize + compression
• v4: Rotation + compression
• v5: Contrast + compression
• v6: Crop + resize + compression

Ground Truth

The filename prefix indicates which original each image was derived from:

• modified_03_brightness.jpg → original_03.jpg
• original_03__rotate6deg__compress__q50__v4.jpg → original_03.jpg

Images in random/ are not derived from any original and should be rejected.

Setup

pip install pillow opencv-python numpy

Usage

Clone this repository and use the images to build and test your forensic matching system:

git clone https://github.com/delveccj/EAS510_Assignment1.git
cd EAS510_Assignment1

Your system should:

1. Register the 10 original images
2. For each test image, determine which original it came from (or reject it)
3. Display transparent reasoning showing how each rule contributed to the decision