DOUBLE JPEG COMPRESSION DETECTION FOR DISTINGUISHABLE BLOCKS IN IMAGES COMPRESSED WITH SAME QUANTIZATION MATRIX

This project performs error analysis on JPEG-compressed images using the UCID dataset. It identifies rounding and truncation errors in DCT coefficients, processes them, and trains a Support Vector Machine (SVM) model for classification of iamges as double or simple compressed.

---

Demo Video Link

📁 Folder Structure

project_root/
├── *.py ``                # Python scripts
├── *.m                     # MATLAB scripts
└── ...

├── data/                   # Expects `ucid.v2` folder here (level equivalent to root)

---

📦 Setup Instructions

1. Dataset Setup

• Download the UCID dataset (e.g., from here).
• Place it outside the project root directory like so:

../data/ucid.v2/

---

2. Install Python Dependencies

Install using:

pip install -r requirements.txt

MATLAB is required for intermediate processing steps.

---

🚀 Running the Pipeline

Step 1: Generate Grayscale Images

python data_maker.py

This converts all UCID images to grayscale JPEGs using a specific quality factor.

---

Step 2: Generate Patches

Run patch_maker.py twice:

python patch_maker.py

• First with train=True
• Then with train=False

Modify the path variable inside the script before each run to ensure correct file I/O.

---

Step 3: MATLAB DCT Error Computation

Run this in terminal:

matlab -nodisplay -nosplash -r "save_error_images; exit"

This will:

• Compute error blocks
• Extract DCT coefficients
• Separate rounding and truncation errors
• Save .mat files for training and testing

---

Step 4: Convert .mat to .npz

python run_all_mat_2_npz.py

Converts MATLAB .mat files to NumPy .npz format.

---

Step 5: Train and Evaluate SVM

python svm.py

Trains an SVM classifier on the error features and evaluates the performance.

---

Step 6: Visualization

To save error blocks as images:

python visualise.py

To generate plots for rounding and truncation error distributions:

python visualise_r_t.py

---

📌 Notes

• MATLAB is required for DCT error extraction.
• Ensure file paths are correctly set in each script before running.
• Tested with Python 3.8+ and MATLAB R2021b.

---

📄 License

This project is intended for academic and research use only.