Signature Forgery & Fraud Detection

Overview

This project implements an AI-based fraud detection system with a specific focus on signature forgery detection. The system processes signature images, extracts key features, and trains a neural network to classify whether a signature is genuine or forged. Additionally, it includes fraud detection mechanisms to identify suspicious activities based on signature analysis.

The model is built using deep learning and computer vision techniques, leveraging image processing and statistical analysis to enhance accuracy.

---

Project Features

✅ Signature Image Preprocessing – Converts images to grayscale and binary format, removes noise, and extracts the signature portion.
✅ Feature Extraction – Extracts critical signature features such as ratio, centroid, eccentricity, solidity, skewness, and kurtosis.
✅ Neural Network Model – A multilayer perceptron (MLP) is trained to classify signatures as genuine or forged.
✅ Fraud Detection – Identifies potential fraud cases by analyzing inconsistencies in signature data.
✅ CSV-based Data Storage – Stores extracted features in CSV format for training and evaluation.
✅ Model Training & Testing – Uses TensorFlow to train and evaluate the fraud detection model.
✅ Performance Evaluation – Measures accuracy, precision, and recall for fraud detection.

---

Code Structure

1️⃣ Libraries Used

• NumPy – Numerical operations
• Pandas – CSV file handling
• Matplotlib – Image visualization
• SciPy & Skimage – Image processing and region properties extraction
• TensorFlow – Deep learning model for signature classification
• OS – File and directory management

---

2️⃣ Functions & Modules

📌 Image Processing & Preprocessing

• rgb2grey(img): Converts RGB images to grayscale.
• grey2bin(img): Converts grayscale images to binary using Otsu’s thresholding.
• preproc(path, img=None, display=True):
  • Converts the image to grayscale & binary.
  • Extracts only the signature part from the image using a bounding box.

📌 Feature Extraction

• Ratio(img): Computes the ratio of white pixels to total pixels.
• Centroid(img): Calculates the signature’s center of mass.
• EccentricitySolidity(img): Extracts signature eccentricity & solidity.
• SkewKurtosis(img): Computes the skewness and kurtosis of the signature.
• getFeatures(path, img=None, display=False): Extracts all features and formats them for analysis.

📌 Data Handling & CSV Generation

• makeCSV(): Generates CSV files with extracted signature features for training & testing.
• getCSVFeatures(path, img=None, display=False): Extracts & stores signature features in CSV format.
• readCSV(train_path, test_path, type2=False): Reads CSV data and prepares it for model training.

📌 Fraud Detection & Signature Classification

• multilayer_perceptron(x): Defines the architecture of the deep learning model.
• trainAndTest(rate=0.001, epochs=1700, neurons=7, display=False):
  • Trains the neural network on signature features.
  • Evaluates model accuracy.
• evaluate(train_path, test_path, type2=False):
  • Loads dataset from CSV files.
  • Trains & evaluates fraud detection accuracy.

---

🛠️ Workflow

1️⃣ Image Preprocessing

✔ Convert the input signature image to grayscale.
✔ Convert grayscale image to binary using Otsu's thresholding.
✔ Extract only the signature portion using bounding box detection.

2️⃣ Feature Extraction

✔ Extract key features:

• Signature ratio, centroid, eccentricity, solidity, skewness, kurtosis.
  ✔ Store the extracted features in CSV files for training & testing.

3️⃣ Model Training & Fraud Detection

✔ Train a neural network (MLP) on extracted signature features.
✔ Evaluate performance using accuracy, precision, and recall metrics.
✔ Detect fraudulent signatures using statistical analysis & deep learning predictions.

---

📌 Usage

Training the Model

1️⃣ Prepare Training Data

• Store genuine and forged signatures in real/ and forged/ folders.
• Run makeCSV() to generate training & testing feature datasets.

2️⃣ Train the Model

train_accuracy, test_accuracy, evaluation_time = trainAndTest()

• This function will train & evaluate the fraud detection model.
• Outputs accuracy & evaluation time.

Testing a Signature

testing("path/to/test/signature.jpg")

• Extracts features from a test signature & saves it in a CSV file.

Evaluate a Custom Signature

train_person_id = input("Enter person's ID: ")
test_image_path = input("Enter path of signature image: ")

train_path = f'Features/Training/training_{train_person_id}.csv'
testing(test_image_path)
test_path = 'TestFeatures/testcsv.csv'

train_accuracy, test_accuracy, evaluation_time = trainAndTest()
print(f"Training Accuracy: {train_accuracy}")
print(f"Testing Accuracy: {test_accuracy}")
print(f"Evaluation Time: {evaluation_time}")

• Trains the model on a custom dataset.
• Evaluates performance on a new test signature.

---

🔍 Example Output

Training Accuracy: 98.5%
Testing Accuracy: 96.2%
Time taken for evaluation: 2.5s

• The model successfully classifies signatures with high accuracy.

---

🚀 Customization & Enhancements

• Modify the neural network layers & hyperparameters to improve accuracy.
• Integrate real-time signature verification for banking & financial applications.
• Expand the fraud detection model to detect anomalies in financial transactions.

---

📝 Notes

• The project is designed using TensorFlow v1 (tf.compat.v1).
• Consider upgrading to TensorFlow v2 for improved performance.

---

🌟 Conclusion

This AI-powered Signature Forgery & Fraud Detection System provides an effective way to classify signatures and detect fraud. The model can be customized & optimized for financial security applications, legal document verification, and banking fraud detection.