A Clustering-based Offline Signature Verification System for Managing Lecture Attendance

Attendance management in the classroom is important because in many educational institutions, sufficient number of class attendance is a requirement for earning a regular grade in a course. Automatic signature verification is an active research area from both scientific and commercial points of view as signatures are the most legally and socially acceptable means of identification and authorization of an individual. Different approaches have been developed to achieve accurate verification of signatures. This paper proposes a novel automatic lecture attendance verification system based on unsupervised learning. Here, lecture attendance verification is addressed as an offline signature verification problem since signatures are recorded offline on lecture attendance sheets. The system involved three major phases: preprocessing, feature extraction and verification phases. In the feature extraction phase, a novel set of features based on distribution of black pixels along columns of signatures images is also proposed. A mean square error of 0.96 was achieved when the system was used to predict the number of times students attended lectures for a given course.