FACE RECOGNITION USING LOCAL TETRA PATTERN

Abstract

Face detection remains a critical task in computer vision, serving as a foundational step for various applications, including facial recognition, surveillance, emotion analysis, and human-computer interaction. Traditional texture descriptors like Local Binary Pattern (LBP) and Local Ternary Pattern (LTP) have been widely used due to their computational efficiency and simplicity. However, their effectiveness significantly degrades in the presence of varying illumination, noise, and non-uniform facial textures. To overcome these limitations, this project explores the use of Local Tetra Pattern (LTrP) and its advanced derivatives as robust descriptors for facial texture representation and detection. Local Tetra Pattern extends the capabilities of traditional binary and ternary patterns by encoding directional edge responses, allowing for more discriminative representation of facial features. In this study, we implement and evaluate the LTrP-based descriptor across multiple benchmark datasets including ORL, JAFFE, and FERET. The project presents a comprehensive end-to-end pipeline—ranging from face localization using the Viola–Jones algorithm to histogram-based feature extraction and support vector machine (SVM) classification. Our results demonstrate that LTrP  outperform classical texture descriptors in terms of recognition accuracy, robustness to illumination variations, and computational efficiency. These findings suggest that tetra pattern-based descriptors offer a lightweight yet powerful alternative to deep learning models in scenarios where resources are constrained or interpretability is prioritized.