Comparative Study of HMM Topologies for Signature Verification (Ergodic vs. Left-to-Right)

This research explores the impact of two distinct Hidden Markov Model (HMM) topologies—Ergodic and Left-to-Right—on the performance of online signature verification systems. Using the SVC 2004 dataset and pressure-based hybrid wavelet transform (HWT) features, we systematically evaluate each topology’s classification accuracy, convergence speed, and computational cost. Our experimental framework includes varying the number of HMM states, training samples, and observation symbols to examine how these topologies influence Equal Error Rate (EER), False Acceptance Rate (FAR), and False Rejection Rate (FRR). Results indicate that while Ergodic HMMs provide superior accuracy due to their flexibility, Left-to-Right models converge faster and demand fewer computational resources. This study provides practical recommendations for selecting an HMM topology based on the intended application’s performance and efficiency requirements.