Local derivative pattern versus local binary pattern: face recognition with high-order local pattern descriptor.

Area of Science:

• Computer Vision
• Pattern Recognition
• Biometrics

Background:

• Face recognition systems rely on effective feature descriptors.
• Existing methods like Local Binary Patterns (LBP) have limitations in capturing intricate facial details.
• High-order local patterns offer potential for enhanced feature representation.

Purpose of the Study:

• To propose and evaluate a novel high-order local pattern descriptor, Local Derivative Pattern (LDP), for face recognition.
• To compare the performance of LDP against LBP in face identification and verification tasks.
• To demonstrate the effectiveness of LDP in capturing detailed directional pattern features.

Main Methods:

• Developed the Local Derivative Pattern (LDP) framework, a generalized approach for encoding directional pattern features based on local derivative variations.
• Proposed n(th)-order LDP to encode (n-1)(th)-order local derivative direction variations, capturing richer spatial relationships than LBP.
• Evaluated LDP using both gray-level and Gabor feature images.

Main Results:

• Extensive experiments on multiple benchmark face databases (FERET, CAS-PEAL, CMU-PIE, Extended Yale B, FRGC) were conducted.
• High-order LDP consistently outperformed LBP in both face identification and face verification.
• LDP demonstrated superior performance across various facial conditions and datasets.

Conclusions:

• The proposed high-order Local Derivative Pattern (LDP) is a highly effective descriptor for face recognition.
• LDP's ability to encode detailed local derivative variations leads to significant performance improvements over LBP.
• LDP offers a robust and accurate solution for challenging face identification and verification scenarios.