Authors :

Mohd Hafeez and Chong Wei Ling

Address :

Department of Cloud Computing, Universiti Teknologi MARA, Malaysia

Department of Data Science, Universiti Sains Malaysia, Malaysiae

Abstract :

Handwriting is a way for people to express themselves through the written word. Character design is an art form in which everyone has their distinct flair. Biometrics and security applications rely on handwriting identification and verification. Inconsistencies in stroke velocity, pressure, and handwriting style diversity are common challenges for traditional approaches. Differences in pressure, stroke irregularities, and handwriting style provide problems for conventional approaches. This project proposed a novel method called DRLCOA-HI to improve handwriting identification and verification methods, both online and offline, by merging Deep Reinforcement Learning (DRL) with Capuchin Optimization Algorithms (COA). This approach uses DRL to determine the best ways to make decisions and extract features. Then, it uses Capuchin optimization to get the model's parameters and hyperparameters just right. The architecture is built around a dual-stream convolutional neural network that can extract features from offline and online stroke data. Then, a DRL agent determines the most important discriminative aspects. The DRL agent's policy and value functions are optimized using Capuchin optimization. Compared to current approaches, the results demonstrate substantial improvements in the accuracy of identification and verification. On average, the accuracy of offline identification increases by 8This technology is incredibly versatile and can be used for forensic and security purposes. It shows improved resilience against many sorts of forgeries and variations in digital and physical formats by 2%, and online verification accuracy increases by 7.1% and 6.8%. This technology is incredibly versatile and can be used for forensic and security purposes. It shows improved resilience against many forgeries and variations in digital and physical formats.

Keywords :

Handwriting identification, Deep Reinforcement Learning, Capuchin Optimization Algorithms.