ETLBP and ERDLBP descriptors for efficient facial image retrieval in CBIR systems.

The traditional Local Binary Pattern (LBP) employs a 3x3 pixel window and examines the intensity differences between the center pixel and nearby neighbourhood pixels. However, LBP excludes the magnitude of difference information entirely, which highly enhances the discriminative performance between classes. In this work, we propose two new feature descriptors called Extended Transition-LBP (ETLBP) and Extended Radial Difference-LBP (ERDLBP) that include the mean of the magnitude difference of each neighbourhood pixel from the central pixel. The robustness of the proposed descriptors is investigated on four publicly available facial databases. The study has established the effectiveness of the feature descriptors. The experimental findings show that the suggested methods statistically outperformed the existing state-of-the-art methods. [ABSTRACT FROM AUTHOR]