Multimodal Approach to Writer Identification from Arabic Handwriting

Writer identification is a process of determining the authorship of handwritten texts, which is important for applications in forensics, historical document analysis, and biometric security. Traditional writer identification methods often rely on images of handwritten documents and require extensive preprocessing, limiting their practicality.  Additionally, the handwritten documents do not explicitly reflect hand kinematics features, thus omitting important individual handwriting patterns crucial for writer identification. This study presents a novel approach to writer identification from surface Electromyography (surface EMG) signals, hand, and stylus kinematics of Arabic handwriting. Surface EMG signals, together with hand and stylus kinematic features, were recorded from 50 participants as they wrote a short text sample. Four deep learning models were developed: one trained on hand kinematics, another on stylus kinematics, the third trained on combined hand and stylus kinematics features and the last one solely on surface EMG signals. The hand kinematics–based model achieved the highest identification accuracy (91.32%), followed by the surface EMG model (90.74%, using only three electrodes), and the combined hand and stylus kinematics model (90.71%). All models substantially outperformed the stylus-only model, which achieved an accuracy of 74.10\%. Additionally, interpretable machine learning techniques were used to reveal the most important features and electrodes for writer identification. These results highlight the potential for hand kinematics for robust writer identification, making it feasible to use with traditional handwriting tools, e.g. by tracking hand movements while writing with a standard pen. The proposed approach can be useful for applications in forensic analysis and biometric authentication.