A Seminar Report

on

Multi-Modal Biometric Recognition for Face and Iris using

Gradient Neural Network (Gen-NN)
By

SHASHANK RAJKUMAR RAMTEKE

Roll No:305C014
TE (Computer Engineering)

Under the guidance of

Prof L. B. Pawar

Department of Computer Engineering

Sinhgad College of Engineering,

Vadgaon (Bk.), Pune-411041

Accredited by NAAC A+ Grade

Affiliated to Savitribai Phule Pune University, Pune

2024-2025
 Date:

CERTIFICATE
This is certified that the Seminar Report entitled

Multi-Modal Biometric Recognition for Face and Iris using Gradient Neural
Network (Gen-NN)
Submitted by

SHASHANK RAJKUMAR RAMTEKE

Has successfully completed her Seminar Report and Presentation under the supervision of
Prof L. B. Pawar for the partial fulfilment of Third Year of Bachelor of Engineering,
Computer Engineering of Savitribai Phule Pune University. This work has not been
submitted earlier elsewhere for any degree.

Prof L. B. Pawar Dr. M. P. Wankhade Guide

Seminar Guide Head,
Computer Engineering Department

Dr. S. D. Lokhande,
Principal,
Sinhgad College of Engineering, Pune

Internal Examiner External Examiner

(iv)
 Abstract
The Biometric systems are widely adopted for personal recognition due to their high level of security and
accuracy, especially in fields such as banking, healthcare, and access control. This report presents a multi-
modal biometric recognition system that combines face and iris recognition using Gradient Neural
Networks (Gen-NN). The use of both face and iris biometrics improves recognition accuracy, overcoming
limitations posed by unimodal biometric systems. Pre-trained models such as ResNet-101 and WaveNet
are employed for feature extraction, while the VISA biometric dataset is used for system evaluation. The
report demonstrates that the proposed system achieves significant performance improvements in accuracy,
sensitivity, and specificity.

(iv)
 Acknowledgement

I take this opportunity to express our profound sense of gratitude and respect to all those who
directly or indirectly helped me through the duration of this work. It gives me great pleasure
to present a seminar on Multi-Modal Biometric Recognition for Face and Iris using
Gradient Neural Network (Gen-NN). I am extremely obliged to our seminar guide Prof L.
B. Pawar in the Department of Computer Engineering for helping and providing me with
proper guidance. Timely suggestions made it possible for me to complete this seminar
successfully.

I will fail in my duty if I don't acknowledge a great sense of gratitude to the head of the
computer engineering department Dr. M. P. Wankhade and the entire staff member of our
department for their cooperation.

Shashank Rajkumar Ramteke

(305C014)

(iv)
 List of Figures

Sr. No. Name Page No.

1 Figure 1.1: The benefits of 3

automating AI in cybersecurity

2 Figure 2.2: Application of AI 11

in Cybersecurity

(iv)
Sr. No. Name Page No.

1 Table 4.1: Accuracy for 9

Various Epochs

2 Table 4.2: Sensitivity 10

and Specificity
Comparison

3 Table 4.3: Dataset 10

Statistics

(iv)
 TABLE OF CONTENTS

Title Page
Certificate Page i
Abstract ii
Acknowledgement iii
List Of Figures iv

1 Introduction 1
1.1 Motivation
1.2 Timeline/Evolution

2 Literature Survey 3
2.1 Fundamentals

2.2 Related Work

2.3 Survey Conclusion

3 Methodology 8
3.1 Selected Algorithm & Technology

4 Results and Discussion 11

4.1 Result
4.2 Conclusion of result
5 Conclusion 16
6 References 18

(iv)
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Chapter 1

Introduction

1.1 Motivation

In today's digital world, there is an increasing need for secure authentication systems,
particularly in sensitive areas such as finance, healthcare, and government services.
Traditional password-based authentication systems are prone to security vulnerabilities like
theft or unauthorized access. Biometric systems, which leverage physiological or behavioral
characteristics such as fingerprints, facial features, or iris patterns, provide a more secure
alternative.

While unimodal biometric systems (using only one trait) face challenges like noisy data,
spoof attacks, and intra-class variability, multi-modal biometric systems combine multiple
traits to improve accuracy and security. This report explores the use of a multi-modal
biometric recognition system that combines facial and iris recognition using Gradient Neural
Networks (Gen-NN) for improved accuracy and robustness.

1.2 Timeline/Evolution

Biometric technologies have evolved from basic fingerprint recognition to sophisticated

multi-modal systems that combine multiple biometric traits for enhanced accuracy. Early
biometric systems often relied on a single trait, such as fingerprints or face recognition, but
faced challenges such as environmental noise and spoofing. Multi-modal systems were
developed to address these challenges by combining multiple traits, providing redundancy
and increased reliability.

The integration of machine learning techniques, particularly deep learning models like
CNNs, has further improved the performance of biometric systems. These models enable
large-scale processing of biometric data, leading to real-time applications in areas like airport
security, banking, and medical diagnostics.

1
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Figure 1.1: A Timeline of Biometric Systems Evolution

A chart depicting the evolution from unimodal systems (fingerprint, face recognition) to
modern multi-modal systems (face, iris, and vein recognition) using deep learning models.

1.3 Organization of Report

This report is organized into five sections:

1. Introduction: Provides an overview of the motivation behind multi-modal biometric systems,

the evolution of biometric recognition, and the organization of the report.

2. Literature Survey: Reviews existing biometric recognition systems and algorithms,

emphasizing the need for multi-modal systems.

2
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

3. Methodology: Describes the Gen-NN-based multi-modal biometric system, including its

preprocessing, feature extraction, and classification stages.

4. Results & Discussion: Analyzes the performance of the proposed system compared to other
biometric recognition systems, with a focus on accuracy, sensitivity, and specificity.

5. Conclusion: Summarizes the findings and suggests directions for future research.

3
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Chapter 2

Literature Survey

2.1 Fundamentals of Biometric Recognition

Biometric recognition systems use physiological or behavioral traits to identify individuals.

Common biometric traits include fingerprints, face recognition, and iris recognition. These
systems capture biometric data using sensors and analyze the data using algorithms to verify
the identity of an individual.

Unimodal systems rely on a single biometric trait, but they face several limitations, such as
noisy data, non-universality, and spoofing attacks. Multi-modal biometric systems, which use
multiple traits, help overcome these limitations by providing redundancy and higher accuracy
in decision-making.

Figure 2.1: Biometric Recognition Framework

This figure illustrates a typical biometric recognition system, including data acquisition,
preprocessing, feature extraction, and classification stages for both face and iris data. The
figure also shows the decision-making process in the Gen-NN model.
4
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

2.2 Related Work

Several studies have focused on multi-modal biometric systems that integrate various
biometric traits. Daas et al. (2020) proposed a deep learning-based multi-modal biometric
system using finger vein and finger knuckle print, demonstrating improved performance with
models such as AlexNet and ResNet. Similarly, Wang et al. (2022) developed a CNN-based
biometric recognition system that integrates face and finger vein biometrics, which achieved
higher accuracy than traditional unimodal systems.

Zhou et al. (2020) developed a hybrid fusion model using iris, palm vein, and finger vein
data, showing promising results in terms of accuracy and stability. However, most of these
models face challenges in terms of computational efficiency and scalability, especially when
applied in real-time applications.

2.3 Survey Conclusion

The literature highlights the advantages of multi-modal biometric systems in terms of

accuracy, reliability, and resistance to spoofing attacks. However, the computational
complexity of these systems remains a challenge, particularly when applied to large datasets
and real-time applications. This report aims to address these challenges by proposing a multi-
modal system that combines face and iris recognition using Gen-NN.

5
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Chapter 3
Methodology

3.1 Selected Algorithm: Gradient Neural Network (Gen-NN)

The proposed multi-modal biometric system combines face and iris recognition using a
Gradient Neural Network (Gen-NN). This system integrates two distinct feature extraction
methods—ResNet-101 for facial recognition and WaveNet for iris recognition—to achieve
high accuracy and robustness.

Figure 3.1: Feature Extraction Diagram

This figure presents a detailed diagram of the feature extraction process. It includes facial
feature extraction using ResNet-101 and iris feature extraction using WaveNet, followed by a
fusion process that combines the two sets of features before being passed into the Gen-NN
classifier.

6
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Preprocessing

• Face Preprocessing: Facial images are preprocessed using filtering, geometric normalization,
and illumination normalization techniques. These steps remove noise and improve the quality
of the images.

• Iris Preprocessing: The iris images are preprocessed through background removal,
normalization, and contrast enhancement to optimize the feature extraction process.

Feature Extraction

• ResNet-101 (Face): ResNet-101 is a deep convolutional neural network that extracts key
facial features using a 101-layer architecture. This model has been trained on large datasets
and is capable of extracting highly discriminative facial features.

• WaveNet (Iris): WaveNet is used for iris feature extraction, employing Gabor wavelets and
Discrete Wavelet Transform (DWT) to capture detailed iris patterns. The combination of these
methods provides accurate and efficient feature extraction.

3.2 Mathematical Model

The Gen-NN model uses backpropagation to minimize the classification error. For a given
input feature vector XXX, the model computes the output Y as:

Y=f(WX+B)

Where:

• W is the weight matrix,

• X is the input feature vector,

• B is the bias,

• f is the activation function.

7
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

The model adapts dynamically to minimize the classification error, improving accuracy
as the system is trained on the dataset.

8
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Chapter 4
Results and Discussion

4.1 Performance Comparison

The Gen-NN model was evaluated using the VISA dataset, which contains face and iris
images. The performance of the system was compared with other classifiers, such as
Random Forest (RF), Decision Tree (DT), and CNN, based on accuracy, sensitivity, and
specificity.

Table 4.1: Accuracy for Various Epochs

Epoch Gen-NN RF DT CNN

Count Accuracy Accuracy Accuracy Accuracy
(%) (%) (%) (%)

100 75.34 62.18 63.57 64.05

200 80.86 81.91 86.78 88.05

300 82.01 88.36 88.98 89.78

500 93.77 N/A N/A N/A

9
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

4.2 Sensitivity and Specificity

The system’s sensitivity and specificity were compared across various algorithms. The
Gen-NN model demonstrated significantly higher sensitivity and specificity than the other
classifiers, indicating that it is more capable of correctly identifying true positives and
minimizing false positives.

Table 4.2: Sensitivity and Specificity Comparison

Classifier Sensitivity Specificity (%)

Gen-NN 91.82 91.99

RF 79.81 81.21

DT 82.85 84.74

CNN 88.65 87.47

Table 4.3: Dataset Statistics

Trait Number of Classes Modality

Images

Face 5000 100 Visual

Iris 4821 90 Infrared

Combined 9821 N/A Mixed

10
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Chapter 5

Conclusion

The multi-modal biometric recognition system proposed in this report combines face and iris
recognition using Gradient Neural Networks (Gen-NN). The use of pre-trained models such
as ResNet-101 and WaveNet improves feature extraction, leading to significant
improvements in accuracy, sensitivity, and specificity. The evaluation of the system on the
VISA dataset shows that the Gen-NN model outperforms traditional classifiers like Random
Forest and Decision Tree in terms of recognition performance. Future work could focus on
optimizing the system for real-time applications and reducing computational complexity.

11
TE,COMPUTER(SCOE),PUNE
 Mul�-Modal Biometric Recogni�on for Face and Iris using
Gradient Neural Network (Gen-NN)

Chapter 6
References

1 Daas S., Yahi A., et al. (2020). Multimodal Biometric Recognition Systems using Deep
Learning.

2 Wang Y., Shi D. (2022). CNN Approach for Face and Finger Vein Fusion.

3 Zhou C., Huang J., et al. (2020). A Hybrid Fusion Model of Iris, Palm Vein, and Finger
Vein

4 Min R., Xu S., Cui Z., "Single-sample face recognition based on feature expansion", IEEE
Access, 2019

5 Ouyang A., Liu Y., Pei S., Peng X., He M., Wang Q., "A hybrid improved kernel LDA and
PNN algorithm for efficient face recognition", Neurocomputing, 2020

6 Alay N., Al-Baity H.H., "Deep Learning Approach for Multimodal Biometric Recognition
System Based on Fusion of Iris, Face, and Finger Vein Traits", Sensors, 2020(MDPI)

7 Simonyan K., Zisserman A., "Very deep convolutional networks for large-scale image
recognition", arXiv, 2014(SpringerLink).

8 Zeghina A.O., Zoubia O., Behloul A., "Face Recognition Based on Harris Detector and
Convolutional Neural Networks", International Symposium on Modelling and
Implementation of Complex Systems, 2020

12
TE,COMPUTER(SCOE),PUNE