Revolutionizing facial image retrieval: Multi-block and mean based local binary patterns with sign and magnitude analysis

Abstract

Robust and accurate approaches are in high demand in the field of facial image retrieval systems. The current methods are not as resilient overall since they mostly rely on sign information within small 3 × 3 or 5 × 5 pixel windows. We provide a novel local binary descriptor specifically designed for facial image retrieval, called Multi-scale Block and Mean-based Local Binary Pattern (MBM-LBP), to address this issue head-on. By utilizing a larger 6 × 6 pixel window and taking into account the sign and magnitude of nearby pixels holistically, MBM-LBP represents a paradigm leap in system robustness and improves the richness of feature representation. The suggested MBM-LBP is carefully examined by means of thorough evaluations using two face image datasets. The results clearly demonstrate MBM-LBP’s superiority over current state-of-the-art methods in the field of face image retrieval. In addition to improving retrieval accuracy, MBM-LBP has the potential to provide more accurate and consistent results for a broad range of real-world uses. This ground-breaking invention paves the way for improved face image retrieval systems, catering to the diverse requirements of multiple industries where reliable and effective retrieval is vital. Facial image retrieval is about to enter a new era marked by significant improvements in both performance and utility, thanks to the leadership of MBM-LBP.

Keywords

• CBIR
• LBP
• 6 × 6 MB-LBP
• Local feature extraction
• Texture descriptor