Automatic Identification and Recognition System Using Face in A Group For Student Attendance

Abstract

In this thesis, we propose a narrative face detection and identification system in a group for class attendance. This system is based on features of face. The importance of this research is, we address the issue of identification of student in the group for class, selection of the best identification and detection techniques by investigating different types of feature extraction techniques and classification methods with different databases of face through confirmatory as well as negative experiments. Further, this identification technique is incredibly abused for the constancy of validation and security by means of investigation of picture and computerization. newlineThe proposed research work is divided in two parts; the first approach is detection of group faces, and the second approach is identification of detected faces for matching purpose from trained dataset. Here we used three datasets, SEAS-FR database and IMF database are available online and VCAGF is our regional database. For face detection purpose, Viola Jones detector is used. HOG and LBP feature extraction methods applied for identification system with different types of binary classifiers like Random Forest (RF), Naïve Bayes (NB), K- Nearest Neighbor (KNN), Support Vector Machine (SVM) and Neural Network. Then passing them into the performance measures and evaluating the identification system. The performance of proposed system for face identification in a group is evaluated in terms of Accuracy (ACC), True Positive Rate (TPR), Specificity (SPC), Positive Predictive Value (PPV), F1 Score, Matthews Correlation Coefficient (MCC), Negative Predictive Value (NPV), False Acceptance Rate (FAR), False Reject Rate (FAR), and Error rate (ERR). newlineThe first model is group faces identification approach using HOG technique. The experimental result shows significantly improve the correctness of identification coordination in the group for class attendance. The accuracy rate improved up to 100% for SEAS-FR and 98.99% for VCAGF dataset with this approach. The feasibility