Local Binary Pattern (LBP) has been widely used in face recognition for its simplicity, but is very sensitive to noise and relies on bilinear interpolation for non-integer neighbor positions. The Binary Rotation Invariant and Noise Tolerant (BRINT) descriptor improved upon LBP through arc-segment averaging, yet still depends on circular sampling requiring interpolation. This paper proposes the Diamond Sampling Structure-Based Local Adaptive Binary Pattern (DLABP) for face recognition. DLABP introduces three contributions: (1) a diamond sampling structure placing all neighbors at integer grid positions, eliminating interpolation entirely; (2) an average method along the radial direction for noise robustness; and (3) a locally adaptive threshold that recovers noise-corrupted nonuniform patterns. DLABP produces a compact 200-dimensional feature and outperforms LBP and BRINT under both noise-free and noisy conditions.
Neelima et al. (Fri,) studied this question.