MatData: Face Recognition Based Smart Voter ID Registration and Duplicate Detection Framework Using Deep Learning

Abstract

Duplicate and fraudulent voter registrations threaten democratic integrity, enabling illegal voting and compromising democratic processes. Traditional voter registration systems rely on manual document verification and demographic matching, which are vulnerable to identity fraud and cannot reliably detect duplicate identities. Recent advancements in deep learning-based face recognition provide an effective and scalable solution for automated identity verification. This paper presents MatData, an advanced deep learning-based biometric voter registration and duplicate detection framework built upon the DeepFace Facenet512 model. The system captures facial images in real time, generates normalized 512-dimensional embeddings, and performs cosine similarity comparison against a secure embedding database. When similarity exceeds a calibrated threshold, duplicate registration is automatically prevented. The framework integrates authentication, face acquisition, embedding extraction, database management, and automated EPIC generation into a unified scalable architecture. Comprehensive case study evaluation demonstrates improved reliability, reduced redundancy, and strong resistance against identity replication attempts.

Introduction

The paper addresses the issue of duplicate voter registrations, which enable illegal voting and undermine electoral transparency. Traditional voter registration systems rely on manual document verification and demographic matching, which are prone to:

• Human error

• Identity fraud

• Inability to detect duplicate identities reliably

To overcome these limitations, the paper proposes MatData, a face recognition–based voter registration and duplicate detection framework powered by deep learning.

Motivation

With advancements in deep learning and Convolutional Neural Networks (CNNs):

• Human faces can be converted into mathematical representations called embeddings.

• Embeddings uniquely represent individuals in high-dimensional feature space.

• Similar faces produce embeddings with small distance.

State-of-the-art models such as:

• FaceNet

• DeepFace

have demonstrated high accuracy in face recognition tasks.

This motivates integrating embedding-based biometric verification into voter registration systems.

Research Gap

Existing voter systems:

• Rely mainly on demographic data

• Fail to reliably detect duplicate identities

• Lack biometric verification integration

Limited research combines deep learning–based face recognition directly with voter registration systems.

Contributions of MatData

The paper proposes a system with the following key contributions:

1. Real-time face recognition voter registration

2. Automated duplicate detection using facial embeddings

3. Secure embedding database storage

4. Automatic EPIC (Electors Photo Identity Card) ID generation

Proposed Methodology

1. System Architecture

MatData integrates six modules:

• Authentication Module

• Face Capture Module

• Recognition Module

• Duplicate Detection Module

• Database Module

• EPIC Generation Module

The system operates in real time.

2. Face Capture Module

• Uses webcam input

• Employs OpenCV for:

  • Face detection

  • Alignment

• Prepares image for embedding extraction

3. Face Recognition Module

Uses DeepFace Facenet512 model.

Embedding generation:

E = f(I)

Where:

• I = input face image

• E = 512-dimensional embedding vector

The embedding represents the individual in feature space.

4. Duplicate Detection Module

Duplicate detection uses cosine similarity:

Similarity = (E? • E?) / (||E?|| × ||E?||)

Threshold: 0.7

• If similarity > 0.7 → Duplicate detected

• Else → New voter registered

This allows automated comparison against stored embeddings.

5. EPIC Generation Module

• Unique EPIC ID generated using secure hashing

• Issued only after duplicate verification passes

System Architecture & Tools

• SQLite: Stores embeddings and voter metadata securely

• Streamlit: Provides user interface

• Real-time operation (not dataset-based offline evaluation)

Case Study Evaluation

Case Study 1: New Voter Registration

Workflow:

1. Face capture

2. Embedding generation

3. Database comparison

4. EPIC ID issuance

Result: Successful registration.

Case Study 2: Duplicate Detection

When a previously registered individual attempted registration:

• Similarity score exceeded threshold (0.7)

• System flagged duplicate

• Registration prevented

Functional Performance

All modules performed successfully:

• Face Capture ?

• Embedding Generation ?

• Duplicate Detection ?

• EPIC Generation ?

• Database Storage ?

Discussion

Advantages:

• Embedding-based biometric verification improves reliability.

• Automated detection reduces human error.

• Real-time processing enhances usability.

Limitations:

• Sensitive to extreme lighting conditions.

• May struggle with facial occlusion.

• Threshold tuning required for optimal performance.

Conclusion

This paper presented MatData, a face recognition-based voter registration and duplicate detection system. MatData strengthens election security through automated deep learning-based duplicate detection.

References

[1] F. Schroff et al., \'FaceNet: A Unified Embedding for Face Recognition and Clustering,\' CVPR, 2015. [2] Y. Taigman et al., \'DeepFace: Closing the Gap to Human-Level Performance in Face Verification,\' CVPR, 2014. [3] O. Parkhi et al., \'Deep Face Recognition,\' BMVC, 2015. [4] B. Amos et al., \'OpenFace: A General-Purpose Face Recognition Library,\' CMU, 2016. [5] G. Bradski, \'The OpenCV Library,\' Dr. Dobb\'s Journal, 2000. [6] S. Serengil, \'DeepFace Framework,\' 2020. [7] K. Kumar et al., \"Fake Identity Detection using Facial Embeddings,\" Springer, 2022. [8] S. Bhattacharya et al., \"Identity Fraud Detection using Deep Learning,\" IEEE, 2022. [9] I. Kemelmacher-Shlizerman et al., \"Face Recognition at Scale: MegaFace Benchmark,\" CVPR, 2016. [10] A. Jain et al., \"AI and Biometrics for Identity Fraud Detection,\" IEEE, 2021. [11] O. Ardizzone et al., \"Detection of Cross-Document Identity Fraud,\" Springer, 2021. [12] T. Schlett et al., \"Double Trouble? Duplicates in Face Datasets,\" arXiv, 2024. [13] Y. Li et al., \"Real-time Face Recognition in Low Power Devices,\" IEEE IoT, 2022. [14] A. Raj et al., \"Face Recognition Bias in India: Study,\" ACM FAT, 2020. [15] Z. Zhang et al., \"A Survey of Face Anti-Spoofing,\" ACM Computing Surveys, 2020.

Copyright

Copyright © 2026 Devansh Mutha, Mrs. Khushbu Jain, Piyush Chordiya, Yash Jain, Srujan Vyavahare. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.