👁️ I-SOCIAL-DB

A Labeled Database of Images Collected from Websites and Social Media for Iris Recognition

Demonstration scripts and utilities for the 2021 Image and Vision Computing paper
I-SOCIAL-DB: A labeled database of images collected from websites and social media for iris recognition

---

🌐 Overview

I-SOCIAL-DB is a dataset and evaluation resource designed to support research on iris recognition from unconstrained images collected from websites and social media.

Unlike traditional iris datasets acquired with controlled sensors, I-SOCIAL-DB focuses on ocular regions extracted from high-resolution face images available in less controlled, real-world-like scenarios.

The database includes manually labeled information useful for iris segmentation and recognition research, including:

• 👁️ ocular-region images
• 🎯 iris boundary annotations
• 🧩 pixelwise segmentation masks
• ✨ occlusion and reflection labels
• 📊 testing protocols and benchmark material

---

🧠 Research Motivation

Images uploaded to websites and social platforms are often high resolution and may contain enough detail to analyze biometric traits beyond the face. I-SOCIAL-DB was created to help study whether iris recognition can be performed on this kind of imagery, where acquisition conditions are more challenging than in classical biometric datasets.

High-resolution face images
          │
          ▼
Ocular region extraction
          │
          ▼
Manual iris annotation
          │
          ├── Inner iris boundary
          ├── Outer iris boundary
          ├── Iris mask
          ├── Occlusions
          └── Reflections
          │
          ▼
Segmentation and recognition evaluation

---

📁 Repository Structure

I-SOCIAL-DB/
│
├── README.md
│
└── code/
    ├── plot_I_SOCIAL.m          # Plot iris image, segmentation mask, and iris boundaries
    ├── convert_JP2_to_BMP.m     # MATLAB JPEG2000-to-BMP conversion utility
    └── convert_JP2_to_BMP.py    # Python JPEG2000-to-BMP conversion utility

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/AngeloUNIMI/I-SOCIAL-DB.git
cd I-SOCIAL-DB

2. Download the dataset

The dataset is distributed through the official project page:

http://iebil.di.unimi.it/ISocialDB/index.html

Place the downloaded data according to the paths expected by the scripts, or update the paths inside the scripts before running them.

---

🖼️ Plot Iris Annotations

The MATLAB script:

code/plot_I_SOCIAL.m

can be used to visualize:

• the ocular image
• the manually segmented iris mask
• the circles approximating the inner and outer iris boundaries

Run it from MATLAB after setting the correct dataset paths:

cd code
plot_I_SOCIAL

---

🔁 Convert JPEG2000 Images to BMP

The repository includes both MATLAB and Python conversion utilities.

MATLAB

cd code
convert_JP2_to_BMP

Python

cd code
python convert_JP2_to_BMP.py

These scripts are useful when working with environments or algorithms that expect images in BMP format.

---

📊 Dataset Highlights

According to the project description, I-SOCIAL-DB contains:

Item	Description
Ocular regions	Extracted from high-resolution face images
Individuals	400 subjects
Face images	1,643 high-resolution images
Ocular samples	3,286 ocular regions
Labels	Human-expert manual annotations
Annotations	Iris boundaries, masks, occlusions, and reflections

---

🧪 Suggested Use Cases

I-SOCIAL-DB can be used for research on:

• iris segmentation in unconstrained imagery
• iris recognition from social-media-style images
• biometric recognition under non-ideal acquisition conditions
• ocular-region analysis
• robustness testing of segmentation and recognition algorithms
• comparison of controlled vs. uncontrolled iris recognition scenarios

---

📌 Main Files

File	Purpose
plot_I_SOCIAL.m	Displays iris images, segmentation masks, and circular iris boundary annotations
convert_JP2_to_BMP.m	Converts JPEG2000 images to BMP using MATLAB
convert_JP2_to_BMP.py	Converts JPEG2000 images to BMP using Python

---

📚 Paper

If you use this dataset or the related scripts, please cite:

@Article{imavis20,
  author  = {R. {Donida Labati} and A. Genovese and V. Piuri and F. Scotti and S. Vishwakarma},
  title   = {I-SOCIAL-DB: A labeled database of images collected from websites and social media for iris recognition},
  journal = {Image and Vision Computing},
  volume  = {105},
  number  = {104058},
  pages   = {1--9},
  month   = {January},
  year    = {2021},
  note    = {0262-8856}
}

Paper:

https://www.sciencedirect.com/science/article/pii/S0262885620301906

Project page:

http://iebil.di.unimi.it/ISocialDB/index.html

---

👥 Authors

• R. Donida Labati
• Angelo Genovese
• Vincenzo Piuri
• Fabio Scotti
• S. Vishwakarma

Department of Computer Science
Università degli Studi di Milano, Italy

---

🏛️ Acknowledgment

This repository provides demonstration scripts associated with the I-SOCIAL-DB research work. Please refer to the official project page and paper for dataset access conditions, benchmark details, and citation requirements.