✍️ Fake Handwritten Signature Detection using Siamese CNN

A deep learning-based signature verification system designed to distinguish genuine signatures from forged signatures using a Siamese Convolutional Neural Network (Siamese CNN). The system learns similarity patterns between pairs of signatures and provides an automated solution for fraud detection, identity verification, and document authentication.

---

📖 Overview

Handwritten signatures remain one of the most widely used methods for authentication in banking, legal documentation, and digital identity verification. Manual verification is time-consuming and prone to human error.

This project automates signature verification using deep learning and image processing techniques. By leveraging a Siamese CNN architecture, the system compares pairs of signatures and determines whether they belong to the same individual based on learned feature embeddings and similarity scores.

---

🚀 Features

🔍 Signature Verification

• Genuine vs forged signature detection
• Pairwise signature comparison
• Similarity-based authentication
• Writer-independent verification

🧠 Deep Learning

• Siamese CNN architecture with shared weights
• Feature embedding generation
• Distance-based similarity learning
• Binary similarity prediction

🖼️ Image Processing

• Grayscale conversion
• Image resizing
• Image normalization
• Noise reduction
• Data augmentation

🌐 Web Application

• Flask-based interface
• User registration
• Genuine signature upload
• Signature verification portal
• Real-time prediction results

🛡️ Security Applications

• Fraud detection
• Banking authentication
• Legal document verification
• Digital identity verification

---

🏗️ System Architecture

Signature A               Signature B
     │                         │
     ▼                         ▼
 Shared CNN Network     Shared CNN Network
     │                         │
     ▼                         ▼
 Feature Embedding      Feature Embedding
            │
            ▼
     Distance Function
            │
            ▼
      Similarity Score
            │
            ▼
     Genuine / Forged

---

⚙️ Methodology

1. Data Preprocessing

• Grayscale conversion
• Image resizing
• Normalization
• Noise removal
• Data augmentation

2. Feature Learning

• Siamese CNN extracts signature embeddings
• Shared-weight architecture ensures consistent learning

3. Similarity Computation

• Computes distance between embeddings
• Generates similarity score

4. Verification

• Similarity score is evaluated

• Signature classified as:

  • Genuine
  • Forged

---

📊 Results

Model	Accuracy
CNN	53.00%
Siamese CNN	87.31%

Key Outcomes

• Achieved 87.31% verification accuracy using Siamese CNN.
• Significantly outperformed standard CNN architecture.
• Successfully differentiated genuine and forged signatures.
• Demonstrated effectiveness for authentication systems.

---

🛠️ Tech Stack

Machine Learning

• Python
• TensorFlow
• Keras

Computer Vision

• OpenCV
• NumPy

Data Analysis

• Pandas

Visualization

• Matplotlib
• Seaborn

Backend

• Flask

Frontend

• HTML
• CSS
• JavaScript

---

📚 Dataset

CEDAR Signature Dataset

The project uses the CEDAR Signature Dataset containing:

• Genuine signatures
• Skilled forgeries
• Multiple users
• Signature verification pairs

This dataset is widely used in offline signature verification research.

---

📸 Screenshots

Siamese CNN Architecture

Signature Verification Workflow

Web Application Interface

User Registration

Genuine Signature Detection

Forged Signature Detection

📂 Project Structure

project/
│
├── dataset/
├── preprocessing/
├── models/
├── training/
├── verification/
├── app/
├── docs/
│   ├── siamese-cnn-architecture.png
│   ├── workflow.png
│   ├── home-page.png
│   ├── user-registration.png
│   ├── genuine-signature.png
│   └── forged-signature.png
│
└── README.md

---

🔬 Research Foundation

The project is inspired by research in:

• Siamese Neural Networks
• Signature Verification
• Metric Learning
• Deep Learning-based Authentication

Key References

• Bromley et al. (1993)
• Hafemann et al. (2017)
• SigNet (2017)
• Koch et al. (2015)

---

🔮 Future Improvements

• Vision Transformers (ViT)
• Attention-based verification
• Mobile deployment
• Real-time authentication
• Multi-factor verification systems
• Large-scale user verification

---

⭐ Applications

• Banking Authentication
• Fraud Detection
• Legal Document Verification
• Identity Verification
• Secure Access Systems

---

👩‍💻 Author

Kandula Sri Chandhana B.Tech CSE (Artificial Intelligence & Machine Learning) VNR VJIET