Robust Face Recognition using Multi-Resolution Multi-Threshold Local Binary Patterns

Face recognition is a pivotal area of research, with various methodologies utilizing different types of information to improve recognition rates. This paper presents a novel approach for face recognition using Multi-Resolution Multi-Threshold Local Binary Patterns (MRMT-LBP) that emphasizes texture information extracted from grayscale images compared to color-based techniques. We propose a systematic generation of multiple threshold LBP representations with four distinct resolutions, resulting in a total of seventeen LBP layers, each corresponding to a different resolution. These layers are then employed to train a face recognition model. The model is constructed by first identifying the LBP layer that achieves the highest recognition rate, which serves as the first channel of our model. Subsequently, additional LBP layers are systematically integrated to form a second channel, with the best complementary layer selected based on its contribution to recognition performance. This iterative process continues until a decline in the recognition rate is observed, at which point the model is built. Comparative evaluations demonstrate that our approach not only achieves superior recognition rates compared to existing grayscale-based face recognition methods but also outperforms prominent color image-based techniques such as RGB and MCF color models. The results underscore the significance of leveraging grayscale images, revealing that the rich texture information they hold can be effectively enhanced to improve face recognition accuracy.