Weighted statistical binary patterns for facial feature representation

• DOI: 10.1007/s10489-021-02477-1
• Publication Date: 2021-05-31T16:31:33Z
• Authors: Hung Phuoc Truong; Thanh Phuong Nguyen; Yong-Guk Kim
• Keywords: Local Binary Patterns; Feature (linguistics); Representation

Abstract We present a novel framework for efficient and robust facial feature representation based upon Local Binary Pattern (LBP), called Weighted Statistical Pattern, wherein the descriptors utilize straight-line topology along with different directions. The input image is initially divided into mean variance moments. A new moment, which contains distinctive features, prepared by extracting root k -th. Then, when Sign Magnitude components four directions using moment are constructed, weighting approach according to applied each component. Finally, weighted histograms of concatenated build histogram Complementary LBP comprehensive evaluation six public face datasets suggests that outperforms state-of-the-art methods achieves 98.51% ORL, 98.72% YALE, 98.83% Caltech, 99.52% AR, 94.78% FERET, 99.07% KDEF in terms accuracy, respectively. influence color spaces issue degraded images also analyzed our descriptors. Such result theoretical underpinning confirms against noise, illumination variation, diverse expressions, head poses.