Scribd
Upload
10 views24 pages

Final Review

Uploaded by

Amrutha
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
10 views24 pages

Final Review

Uploaded by

Amrutha
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Detecting Fake Face Images in

Biometric Systems using LBP


Feature Extraction and CNN
Under the Supervision of
N. Gowri Sree Lakshmi Asst. Prof
Department of CSE
Presented By:
20551A05D9 – K. Chaturya
20551A0522 – K. Sobharani
20551A0537 – N. Amrutha
20551A0560 – N. Shalem Madhukar
CONTENTS
1) Abstract
2) Objectives
3) Introduction
4) Working of Face Spoofing Detection
5) Existing System
6) Proposed System
7) Dataset
8) Workflow
9) Local Binary Pattern
10) Convolutional Neural Network
11) Feature comparison of real and spoof images
12) Comparison of Accuracy
13) Results
14) Conclusion
15) References
ABSTRACT
“Our project focuses on the development of an advanced
face spoofing detection system using Local Binary
Patterns(LBP) and Convolutional Neural Networks(CNN).
Face spoofing is a activity when someone tries to bypass a
facial recognition systems by using pre-captured images and
videos of a person’s face. Our project contributes to identify
the most used face spoofing techniques called presentation
or print attack, replay attack where the attacker tries to
bypass the system using authorized user photograph and also
the advancement of secure authentication systems,
enhancing security measures in critical applications such as
access control, financial transactions and surveillance."
OBJECTIVES
➢ Create a comprehensive system that combines Local Binary Pattern(LBP) and
Convolutional Neural Network(CNN) techniques to effectively detect face spoofing
attempts.

➢ Incorporate Local Binary Pattern(LBP) as a key feature extraction method to analyze


the texture patterns of facial images.

➢ Implement Convolutional Neural Network (CNN) architecture to perform deep


learning-based classification on the extracted features.

➢ The system aims to achieve high accuracy making it robust against face spoofing
techniques such as presentation and replay attacks.
INTRODUCTION
➢ Facial Recognition technology is becoming increasingly prevalent in various applications from
unlocking smartphones to securing high-stakes facilities.
➢ However, with the rise of this technology comes the growing threat of face spoofing attempts to
fool facial recognition systems using pre-captured photos and videos.
➢ The project aims to develop a robust Face Spoofing Detection system leveraging the combined
power of LBP and CNN for accurate classification.
➢ The proposed model can efficiently classify between real and two main face spoofing
techniques:
• Presentation Attack: Presentation Attack is a face spoofing technique where attacker tries to
bypass a facial recognition system by placing a printed image of a person in-front of biometric
system.
• Replay Attack: Replay attack is a face spoofing technique where attacker tries to bypass a
facial recognition system by placing a mobile which has pre-captured photograph of a person
in-front of biometric system.
➢ The project not only addresses the current challenges but also paves the way for enhanced
security in the era of widespread facial recognition technology.
WORKING OF FACE SPOOFING DETECTION
EXISTING SYSTEM

➢ Previous researches has primarily focused on employing Convolutional


Neural Networks(CNN) as a robust solution.
➢ Convolutional Neural Networks(CNN) is a deep learning algorithm designed
for image processing tasks, utilizing convolutional layers to automatically
learn hierarchical representation of features from input data.
➢ However a critical analysis reveals certain limitations and gaps that prompt
the exploration of alternative methods.
PROPOSED SYSTEM

➢ The proposed system aims to create LBP based


Convolutional Neural Network which can classify
fake and original images.
➢ In this case LBP is first extracted from the images,
and then CNN is used to train the LBP descriptor
photos.
➢ The use of LBP simplifies the task of CNN for
classification of images thus leads to the effective
solution.
DATASET

Dataset containing Real Dataset containing Fake


Images(1500 images) Images(1950 images)
Workflow
Local Binary Pattern
Before moving on to the working of Local Binary Patterns let’s see the before and
after math of an image which undergoes Local Binary Pattern algorithm.

Image before applying Image after applying


Local Binary Pattern Local Binary Pattern
Working of Local Binary Pattern

LBP Code of centre pixel:1+2+16+64+128=211


Why Local Binary Pattern?

➢ Local Binary Pattern excels at capturing


local texture patterns.
➢ It is robust to variations and noise in
LBP of a real image taken straight
out of camera
texture
➢ Enhances CNN’s ability to detect subtle
texture differences.

LBP of a spoof image where mobile


image is placed in front of camera
Convolutional Neural Network

Convolutional Neural Network is a class of deep neural networks designed for


image processing and pattern recognition.
Key components of CNN include :
• Convolutional Layer: Here we use filters to convolve input data, producing
feature maps. It also referred as feature extraction layer.
• Pooling Layer: Pooling Layer down sample feature maps, reducing spatial
dimensions and computational complexity.
• Fully Connected Layer: Fully connected layers integrate features from previous
layers for classification. There layers connect all neurons, enabling the network
to make predictions based on learned features.
Convolutional Neural Network Architecture

Input
Image

Flattened Layer
Conv Pooling Fake
Conv Pooling

224x224x3 Real
Feature Maps
Max-pooled feature maps
128x128x32
64x64x32

Output Layer

Feature Extraction Fully Connected Layer


128 Neurons
Feature Extraction
Feature extraction is the main phase in CNN. CNN classify the image into
either Real or Spoofed Image based on the features extracted from the image.
Although spoof images look similar to real faces there are some features
which separates real image from the spoofed image. Those are:
• Image Texture: Spoofed images doesn’t have natural textures such as skin
texture, wrinkles, hair details etc.. when compared to real images.
• Reflections: Spoofed images contains more reflections in the image due to
the mobile screen than the real image.
• Noise: Spoofed images contains random variations in brightness or colours
that are not part of the actual image.
• Edges: Spoof images often contains unnatural boundaries such as mobile
screen edges, borders etc…
Comparison Between Features Extracted from Real and Spoof Images

Input Image Reflections Noise Edges


Texture

Real Image

Spoof Image
COMPARISION OF ACCURACY

SYSTEM ACCURACY

EXISTING SYSTEM 91.83%

PROPOSED SYSTEM 96.83%


RESULTS

To load the To upload new


trained model image
in background

To exit from
the
To classify application
whether the
given image is
real or fake
Step-1: Generating Model

Step-2: Uploading New Image


Step-3: Classifying Image

Step-4: Exit application


CONCLUSION
➢ Our project focused on advancing face spoofing detection through the integration of
LBP and CNN.

➢ CNN excelled in learning hierarchical features while LBP enhanced the model’s
ability to capture texture based information crucial for distinguishing between
genuine and spoofed faces.

➢ LBP significantly contributed to the model’s ability to discern subtle variations in


facial textures associated with spoofing attempts.

➢ Comparative analysis revealed a notable reduction in false positives and increased


accuracy in detecting various face spoofing attacks.
REFERENCES

1. Luca Guarnera, Oliver Giudice, Cristina Nastasi, and Sebastiano Battiato,


“Preliminary forensics analysis of deepfake images” ,2020.

2. Zhengzhe Liu, Xiaojuan Qi, Jiaya Jia, and P. Torr, “Global texture enhancement for
fake face detection in the wild”,2020

3. Yuanfang Guo and Xiaochum Cao and Wei Zhang and Rui Wang, “Fake Colorized
Image Detection”, 2018.

You might also like