Over the past three decades, a wide array of face-recognition techniques has been developed, driven by the increasing demand for effective methods to identify human faces in various real-world applications. In this paper, we introduce a novel methodology that integrates the Discrete Cosine Transform (DCT) with an enhanced version of Discriminant Linear Discriminant Analysis (D-LDA) and Neural Networks. Initially, we calculate the eigenvectors along with a new Fisher's criterion using our improved D-LDA algorithm. This innovative approach allows us to refine the process of dimensionality reduction, focusing on maximizing class separability. Following this, we compute the projection vectors derived from the training set, which serve as the foundational input when training the neural networks for the specific task of human identity recognition. To evaluate the effectiveness of our proposed methodology, we conducted experimental tests using the ORL face database, which contains a diverse set of face images. The results indicate that our combined approach demonstrates outstanding performance in accurately recognizing human faces, providing significant improvements over traditional methods. These findings suggest that the integration of DCT, improved D-LDA, and neural networks can enhance the robustness and accuracy of face-recognition systems, making them more viable for real-world applications.
Face recognition, Discrete Cosine Transform, Discriminant Linear Discriminant Analysis