Long Short-Term Memory Utilization for Non-Invertible Transformation Face Biometric Recognition System
Hiba Basim Alwan · 2024
A recognition system is important in artificial intelligence research areas. It is broadly used for identification reasons and in several ways like facial recognition. Recognition systems depend on classification systems to work. Classification is a supervised learning technique. Long short-term memory is a type of supervised learning technique and can be used for classification purposes. Feature selection techniques have been used to improve the classification performance and reduce the training time. The black widow optimization algorithm is considered a kind of feature selection technique. Biometric systems are a particular type of recognition system. It influences the recognition system to accomplish its kernel work in identifying users based on the user’s unique biometric data. Face images are broadly used as a biometric trait. Facial recognition is popular, because of its non-intrusive, technological development, and cameras, nowadays, everywhere. Facial recognition is a common selection for detecting known users in secret zones. In this paper, the black widow optimization algorithm has been used to select the suitable feature. Then, the proposed non-invertible transformation method is performed to secure the selected feature. Finally, long short-term memory has been used as a classifier to classify known and unknown users. Two common datasets have been used to evaluate the proposed system. These datasets are FEI and Georgia Tech datasets. Two common measurements have been used to assess the proposed system. These measurements are the classification accuracy and the equal error rate. The experimental results show that the average classification accuracy was 98.61% for the FEI dataset and 98.21% for the Georgia Tech dataset. The average value for equal error rate was 0.68% for the FEI dataset and 0.86% for the Georgia Tech dataset.