MB-ZZLBP: Multiscale Block ZigZag Local Binary Pattern for Face Recognition

This work presents the LBP variant so-called MB-ZZLBP under illumination and expression variations. In MB-ZZLBP, initially mean is computed for all the square sub-blocks (size 2 × 2) of the 6 × 6 pixel window. After the mean computation, 3 × 3 window (pixel) is produced. Then ZigZag pixels are compared with each other, or in other words, the higher-order pixels are subtracted from the lower-order pixels. Differences of those which produce value ≥0 are allocated the label 1, else 0. After encoding each pixel position (binary pattern) eventually results in MB-ZZLBP transformed image. The MB-ZZLBP image is further divided into 3 × 3 sub-regions for extraction of histograms. The fused histogram is the feature size of MB-ZZLBP. The proposed FR approach attains remarkable results on EYB and Faces94 databases.