<p>LBP and most of its variants extract features from 3 × 3 image patch. Although 3 × 3 patch yields good robustness but limits discriminativity due to usage of only 9 pixels. The persisting local descriptors make different combinations and permutations from 3 × 3 patch to form their respective feature size. Literature reveals that there is less work performed in literature in which higher patches are used for feature extraction. To fill this gap, the proposed work makes use of the 4 × 4 patch for feature extraction which extracts more discriminant features than many methods. Precisely, the proposed work first launches two descriptors so-called 3 Directional Sum Based Binary Pattern (3DSBBP) and 3 Directional Product Based Binary Pattern (3DPBBP). Under both, two descriptors are introduced based on Sign (Sn) and Magnitude (M) components called as [3DSSnBBP, 3DSMBBP] and [3DPSnBBP, 3DPMBBP]. The [3DSSnBBP, 3DSMBBP] and [3DPSnBBP, 3DPMBBP] based features are captured from three distinct directions. The descriptor histograms under 3DSBBP and 3DPBBP are merged separately to build respective feature size. Eventually 3DSBBP and 3DPBBP sizes are joined to form the size of discriminant descriptor 3DBBP. FLDA is exploited for contraction and SVMs and NN are used for classification. On ORL, GT, Faces94, JAFFE and EYB datasets astounding consequences are gained by 3DBBP, which is much higher than many methods. 3DBBP achieves the best accuracy of 100%, 96.85%, 100%, 100% and 90.46% chronologically. In contrast to other methods, these accuracies are well superior.</p>