Project Overview
For my final year project, I built a deep learning-based system to detect forged signatures. Signature forgery is a major concern in banking, legal, and administrative workflows, so I wanted to explore how artificial intelligence can help reduce fraud through automation.
Tools & Technologies Used
Python
Streamlit (for building the web app)
ResNet & VGGNet (for image classification)
TensorFlow / Keras
Dataset: CEDAR Signature Dataset
How It Works
The model takes an image of a signature as input
It processes the image through a pre-trained CNN (ResNet or VGG)
The output is a prediction: Genuine or Forged
The system was trained on labeled data containing both genuine and forged signatures, achieving solid accuracy after tuning hyperparameters and balancing the dataset.
The Interface
Using Streamlit, I created a lightweight interface where users can upload a signature image and get instant results. This made the project more interactive and user-friendly.
Results
Model accuracy: ~95%
Fast prediction time (less than 1s per image)
Smooth deployment with a clean UI
What I Learned
Hands-on experience applying convolutional neural networks to a real-world classification problem
Working with imbalanced datasets and image preprocessing
Building and deploying a functional Streamlit web app
Final Thoughts
This project showed me the practical value of deep learning in security and fraud detection. It also solidified my passion for machine learning and opened the door for more real-world AI projects in my future career.