A new patch-based LBP with adaptive weights for gender classification of human face
Wen‐Shiung Chen, Ren-He Jeng · Journal of the Chinese Institute of Engineers · 2020
Recognition methods based on local binary patterns (LBP) depend on a pre-defined structure, such as pixel-to-pixel or pixel-to-patch comparison, for the feature extraction of picture descriptions without the addition of weights. This paper proceeds on the assumption that patch-based LBP should be based on a pyramid structure to enable the computation of gradients using weight parameters pre-defined by eigen theory. The proposed method, namely Adaptive-Weight Patch-based LBP (AWPLBP), seeks to determine the transformation that identifies sets on the hyper-plane capable of maximizing variance in projections. The transformation and quantization loss are used to train the AWPLBP descriptors to facilitate the efficient feature extraction of descriptions. Finally, a classifier with a support vector machine (SVM) is trained using the AWPLBP features for the gender classification of human faces. The experimental results show that the proposed AWPLBP descriptor may achieve better recognition performances than other LBP methods and is competitive with CNN-based methods in gender classification application.