SIGNATURE VERIFICATION

SYSTEM

Batch No:4
Guided by
V.Sri Pragna Dr. B. Sivaramakrishna
L.Pavan Sai
M.Gowthami
ABSTRACT

This project aims to develop an application which is user friendly for a automatic and
real-time signature verification system by using advanced image processing and
deep learning techniques to analyze and compare the dynamic features
signature .Through the use of sophisticated algorithms, the system extracts unique
signature characteristics, including pressure points, stroke patterns, and timing
information, ensuring a robust and accurate verification process .
INTRODUCTION

Signatures are widely used as a method of personal identification and

verification. Many documents like bank cheques and legal transactions require
signature verification.
Two types of signature verification system like manual and automated.
For manual signature verification,we are facing some challenges.
● Time-Consuming
● Human Error
● Forgery and Fraudulent Activities
To overcome the challenges ,an automated system that reduces the reliance on
manual signature verification, making the process more efficient and less prone
to human error.
Literature Survey
AUTHORS TITLE METHODOLOGY DRAWBACKS

Hsin-Hsiung Kao “An Offline Signature The authors propose an off-line

Creating a greater number
* handwritten
Verification and of forged signatures could
and Che-Yen Wen Forgery signature verification method based on a
deep lead to increase in
Detection Method
prediction accuracy.
Based convolutional neural network, DCNN and
on a Single Known unique local feature extraction approach.
sample It
and an explainable focuses on learning more from forged
Deep
signatures than genuine ones
Learning Approach ”
Kumari .K., Rana “Towards improving In this paper, machine learning classifiers
Traditional machine learning
.S. offline are
classifiers may not scale
Signature verification used to verify the signature using four
image well with large datasets
based Authentication
compared to deep learning
based features. Six classifiers are trained
Using machine learning models, which are designed
and
classifiers”
tested on BHSig260 dataset and the to handle large amounts of
obtained data and complex patterns
Anamika Jain, “Handwritten Signature
The proposed architecture is Reduce chances overfitting
Satish
Verification using shallow very simple but extremely
Kumar Singh without data augmentation and
Convolutional Neural efficient in terms of accuracy. A
,Krishna Pratap also designing a model which
Network” custom shallow convolution
works for skilled forgeries as
Singh neural network is used to
well.
automatically learn the
features of signature from the
provided training data.
George S.
“Hybrid writer-independent– In this study, a hybrid WI–WD Enhance the system accuracy
Eskander,
write system is proposed, as a by employing other features,
Robert Sabourin,
Eric compromise of the two and learn from independent
r-dependent offline signature
approaches. When a user is forgeries, during the WI
Granger
verification system.” enrolled to the system, a WI training. Retuning classifier
classifier is used to verify his
parameters for greater
queries.
accuracy can be explored.
Amruta B. Jagtap, Uses a new kind of network The effectiveness of the SNN
“Verification of genuine and
called siamese network, which heavily relies on the quality
Dattatray D. Sawat, forged offline signatures using
is essentially an amalgamation and quantity of the training
Rajendra S. Hegadi, Siamese Neural Network of two CNN's with identical data. If the dataset is small or
weights and hidden layers. contains noise, the network's
Ravindra S. Hegadi (SNN)” ability to distinguish between
genuine and forged signatures
PROBLEM STATEMENT

Design an application for signature verification system that accurately and efficiently
verifies the authenticity of handwritten signatures. The system should be capable of
distinguishing between genuine and forged signatures while handling variability in
signature styles and maintaining high performance in real-time applications.
MAIN OBJECTIVES OF OUR PROJECT

● Main objective is provide the more accurate result.

● Develop a system that can adapt to variations in individuals'
signature styles over time while maintaining accuracy in verification.
● Objective is to decrease the time required for signature verification
compared to manual methods.
● Improve the efficiency of the signature verification system by using
some advanced techniques.
● Build a mobile app that verifies signatures, ensuring secure and reliable
identity authentication for user-scanned documents.
PROPOSED WORK

● Signature acquisition
● Signature Preprocessing
● Feature Extraction
● Signature verification
SIGNATURE ACQUISITION

The process of collecting and capturing signature data, which can be used for
various purposes, such as authentication, verification, or forensic analysis.These are
images of signatures taken from physical documents.
SIGNATURE PREPROCESSING
Preparing raw signature data for further analysis or verification. This step is crucial in
signature verification systems to ensure that the data is clean, consistent, and
suitable for feature extraction and classification.

Function of pre-processor:

● Image resizing
● Image binarizing
● Image thinning
● Image normalizing
FEATURE EXTRACTION

Ratio: Foreground-to-total pixel ratio, indicating the prominence of the

object.

Centroid: Center of mass of the foreground pixels, useful for object

localization.

Eccentricity: Shape elongation measure.

Solidity: Measure of how compact or spread out the signature is, which can
vary significantly between individuals.

Skewness:This measures the asymmetry in the pixel distribution of the

signature.
SYSTEM ARCHITECTURE
PROPOSED ALGORITHM- MULTILAYER
PERCEPTRON (MLP)

A Multilayer Perceptron (MLP) is a type of artificial neural network used in

deep learning. In signature verification systems, signatures are usually
converted into a numerical format that can be processed by an MLP.
Here,how it works:
An MLP consists of multiple layers:
● Input Layer: Each neuron in this layer corresponds to a feature of the input data.
● Hidden Layers: One or more layers where the actual processing occurs. These layers
apply various transformations to the input data using activation functions The neurons in
these layers learn complex patterns and features from the input data.
● Output Layer: This layer provides the final output of the network.(genuine or forged)
RESULTS
RESULTS
CONCLUSION

The project achieved 56% accuracy in distinguishing real signatures from forgeries,
demonstrating the model's effectiveness in signature verification. While the results
are promising, there is scope for improvement by refining algorithms and using
larger, more diverse datasets. We make this system has potential for practical
applications in fraud detection and authentication.
THANK YOU