REALME: An Approach for Handwritten Signature Verification based on Smart Wrist Sensor

• Published in: Pattern Recognition and Image Analysis (IPRIA), International Conference on pp. 1 - 6
• Main Authors: Taimoor, Muhammad, Butt, Huma, Khadim, Tanveer, Ehatisham-ul-Haq, M., Raheel, Aasim, Arsalan, Aamir
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.11.2020
• Subjects: Accelerometers; biometric; classification; Data acquisition; Error analysis; fake signatures; Feature extraction; Gyroscopes; handwritten signature; Performance evaluation; real signatures; Wrist
• ISSN: 2049-3630
• DOI: 10.1109/INMIC50486.2020.9318184

Handwritten signature verification has a wide range of applications e.g., signing banker cheques or any legal documents. Most of the existing systems for handwritten signature verification uses specialized equipment like a smart pen and a tablet to acquire the signatures. Whereas in our proposed scheme, the signatures are captured on a plain sheet of paper using a low cost commercially available wrist-worn MetaWear smart sensor that is capable of recording motions using accelerometer and gyroscope. Smart sensor data from 10 subjects are acquired while performing real and fake signatures. Machine learning algorithm is trained using a known set of real and fake signatures and then using the test data, the algorithm identifies that whether the signatures of an individual are real or fake. A total of 12 trails are obtained from each user resulting in a total of 120 samples (10 users × 12 trails for each user) for genuine signatures and 120 samples for the fake signatures. Feature extraction is applied in time domain to extract sixteen features. Wrapper method for feature selection selects the optimum subset of features from the extracted features. Real and fake signatures are identified using BayesNet algorithm with an accuracy of 98% with a feature vector length of 5. It is evident from the results that the proposed scheme produces results which are comparable to the state-of-the-art methods available in the literature in terms of accuracy and has superior performance in terms of feature vector length.