22CSE71 – Deep Learning Laboratory
Mini Project
Synopsis
Project Title: AI-Powered Signature Forgery Detector
Sl. # Name of the Student USN
1. Varun Quanth 1GA22CS184
2. Yashwanth S R 1GA22CS192
Introduction:
The AI-Powered Signature Forgery Detector aims to automate this process using
Deep Learning. It employs a Siamese Convolutional Neural Network (CNN)
that compares two signature images and determines whether they belong to the
same person or not. By analyzing signature patterns, strokes, and shapes, the
system can effectively detect forged signatures with high accuracy.
This project demonstrates how artificial intelligence can enhance the security and
reliability of signature verification systems while reducing human effort and error.
Base paper related to the Problem Statement Existing Methodology
Base Paper
Title: Offline Signature Verification using Siamese Convolutional Neural Network
Authors: Hafiz Tayyab Rauf, Muhammad Attique Khan, and others
Journal: IEEE Access, 2020
Summary:
This paper proposes a Siamese CNN-based approach for offline signature
verification. The model learns to extract discriminative features from genuine and
forged signatures using identical convolutional branches that share parameters. The
network measures the similarity between two signatures using Euclidean distance
and classifies them as genuine or forged. Experimental results on benchmark
datasets (such as GPDS and CEDAR) showed high verification accuracy, proving
the efficiency of deep feature-based similarity learning.
Existing Methodology
Traditional signature verification methods are mainly manual or feature-based,
which have several drawbacks:
1. Manual Verification:
o Relies on human observation to match signatures.
o Subjective and prone to human error.
o Not suitable for large-scale automated systems.
2. Feature-Based or Statistical Methods:
o Use handcrafted features like shape, angle, and texture.
o Require pre-processing and expert-designed feature extraction.
o Performance degrades with variations in writing styles or scanning
quality.
These limitations motivated the shift toward Deep Learning-based approaches,
where models automatically learn signature features directly from image data.
The proposed AI-powered system overcomes these drawbacks by using a
Siamese CNN that learns similarity patterns autonomously, eliminating the need
for manual feature engineering or visual verification.
Motivation:
Signature forgery is a growing concern in today’s digital and financial systems,
where signatures are still widely used for authentication. Manual verification is
often slow and unreliable, leading to errors and security risks. This motivated the
need for an automated and intelligent system that can verify signatures
accurately.
Using Deep Learning and a Siamese CNN model, this project aims to make
signature verification faster, more reliable, and free from human bias—enhancing
trust and security in real-world authentication processes.
Problem Statement:
Signature forgery is a major concern in authentication systems, especially in
banking, legal, and institutional sectors. Manual verification of signatures is prone
to human error, inconsistency, and bias, making it unreliable for large-scale or
high-security applications. The need for an automated system that can verify
handwritten signatures accurately and efficiently is therefore essential.
The problem addressed in this project is to develop an AI-based model that can
automatically detect forged signatures by comparing two signature images.
The goal is to create a system that minimizes human intervention, enhances
accuracy, and reduces processing time in signature verification tasks.
Objectives:
The main objectives of the AI-Powered Signature Forgery Detector are:
1. To develop an AI-based system capable of automatically distinguishing
between genuine and forged signatures.
2. To implement a Siamese Convolutional Neural Network (CNN) that
learns the similarity between two signature images.
3. To generate a synthetic signature dataset automatically using
handwriting-style fonts and image augmentations for training.
4. To reduce human effort and error in the process of signature verification
by providing an automated, data-driven solution.
5. To achieve high accuracy and reliability in detecting forgery using deep
learning techniques.
6. To create a simple user interface in Colab that allows users to upload two
signatures and instantly check if they match.
Methodology:
The proposed AI-Powered Signature Forgery Detector uses a Siamese
Convolutional Neural Network (CNN) to identify whether two signatures belong
to the same person or not. The system follows these main steps:
1. Dataset Generation
o A synthetic signature dataset is automatically created using
handwriting-style fonts.
o For each virtual person, multiple genuine and forged signatures are
generated with random variations like rotation, blur, and noise.
2. Data Preprocessing
o All images are converted to grayscale, resized to a fixed dimension,
and normalized.
o Signature pairs are prepared:
Genuine Pair (Same Person) → Label = 1
Forged Pair (Different Persons) → Label = 0
3. Model Design (Siamese CNN)
o Two identical CNN branches share the same weights.
o Each branch extracts deep visual features from an input signature.
o The absolute difference between feature vectors is calculated and
passed to dense
Sl. Student Guide
#
Student Name USN
Signature Signature
1.
2.
3.
Faculty In-charge Remarks:
Status: Approved Rejected Needs Revision
Remarks:
Submission
Date:
Sl. # Faculty Name Faulty Signature
1.
2.