Design and Implementation of a Robust Biometric System Based on Face Recognition

Researchers have found the field of face recognition to be of great intrigue. The driving force behind the immense enthusiasm surrounding this subject stems from the necessity to enhance the precision of numerous real-time applications. Over the course of recent years, an abundance of approaches has been recognized and introduced. The intricate nature of facial visuals, coupled with the considerable variations resulting from diverse influences, renders the task of devising and implementing a robust computational system for object recognition and human face recognition all the more arduous.
Extensive training is frequently required by the computer when utilizing supervised learning, thereby leading to prolonged execution times. In the realm of face recognition, the application of robust preprocessing techniques is imperative to attain a commendable accuracy rate. While approaches that encompass both detection and recognition do exist, it is our belief that the lack of a comprehensive end-to-end system capable of carrying out recognition from any given scene is primarily attributed to the challenges associated with alignment. Oftentimes, facial registration is disregarded, under the assumption that the detector will execute a rough alignment, consequently resulting in suboptimal performance in recognition.
In this research, an improved approach to enhance human face recognition was presented, employing the utilization of local Binary Pattern (LBP), maximal response 8 (MR8), and Support Vector Machine (SVM). A noteworthy contribution of this study involves the creation of a novel feature vector derived from LBP and MR8 methods, extracted from the original training dataset, which is subsequently employed to train the SVM. The two feature extraction methods were applied independently, and their outputs were combined to generate a unified feature vector, which was then fed into the SVM. The utilization of reduced image features resulted in higher face recognition accuracy, while simultaneously reducing computational costs, as evidenced by achievement rates of approximately 97% and 84%. The proposed framework was rigorously tested on two datasets, namely the YALE and AT&T data sets, which served as the benchmark. The results obtained were truly remarkable, demonstrating significant improvements in accuracy.