Enhancing Siamese Neural Network Performance with Custom Local Binary Patterns for Face Matching

In this study, we investigate the efficacy of incorporating custom local binary patterns (LBP) as a preprocessing technique to enhance the performance of Siamese neural networks for face matching tasks. Our research focuses on training Siamese models using the DC Face dataset and comparing their performance when trained with and without LBP integration. The custom LBP function is designed to capture fine-grained texture details from facial images, providing robust feature representations that are invariant to variations in lighting conditions and facial expressions. Through rigorous experimentation and evaluation, we analyze the impact of LBP on the threshold point for image differentiation and compare the matching accuracy of Siamese models trained with and without LBP. Our findings reveal significant improvements in matching accuracy and robustness when utilizing the texture features extracted by the custom LBP function. By encoding local pixel structures and capturing subtle facial characteristics, LBP-enhanced Siamese models demonstrate enhanced performance in accurately matching faces across different images. Further-more, our study highlights the importance of using domain-specific preprocessing techniques, such as LBP, to optimize the capabilities of Siamese neural networks for face matching tasks. The results underscore the effectiveness of customizing feature extraction methods to suit the characteristics of the dataset, leading to improved matching accuracy and robustness. Overall, our research contributes valuable insights into the role of LBP in enhancing the performance of Siamese neural networks for image-based similarity tasks, with implications for various real-world applications in face recognition and biometric authentication.