A Hybrid Method of Feature Extraction for Signatures Verification Using Cnn and Hog a Multi-Classification Approach

Abstract

The offline signature verification systems feature extraction is crucial to their performance. The amount and precision of extracted characteristics influence how successfully these algorithms can distinguish authentic and fraudulent signatures. Using a CNN and a Histogram of Oriented Gradients, we established a novel technique to extract features from signature photos. We then selected the most significant attributes using Decision Trees. Integrating CNN and HOG was the last step. Three models—long short-term memory, support vector machine, and K-nearest Neighbor—tested the combination technique. Our approach accurately forecasted the future and utilized the CEDAR information effectively, according to the trials. Since we tested sophisticated false signatures, which are harder to recognize than simple or opposite signs, this accuracy is crucial. The project now includes a Voting Classifier for Dataset Analysis and Feature Extraction. We achieved 100% accuracy for improved Signature Verification utilizing CNN and HOG, a multi-classification approach. Users can easily sign up and log in for testing using a simple Flask framework that uses SQLite, ensuring that the application can be used safely in real life.