A HYBRID METHOD OF FEATURE EXTRACTION FOR SIGNATURES VERIFICATION USING CNN AND HOG A MULTI-CLASSIFICATION APPROACH

Abstract

The quantity and calibration of the extracted features determine how well these systems can distinguish between authentic and fake signatures, making the feature extraction phase of an offline signature verification system critical to the overall performance of these systems.The study presents a hybrid technique for offline signature verification systems feature extraction from signature pictures. To find important features, the approach combines the usage of Convolutional Neural Network (CNN) and Histogram of Oriented Gradients (HOG) approaches. A decision tree feature selection algorithm then comes into play. The hybrid approach was tested on two datasets (UTSig and CEDAR) and assessed using three classifiers: K-nearest Neighbor, support vector machine, and long short-term memory. Even for expert forgeries, the testing findings demonstrated a high degree of accuracy in identifying genuine from fabricated signatures.