Abstract：To solve the problems that the random subspace method had poor robustness to local variation of human face images, high computational complexity using multiple random sampling and fixed sampling rate, a human face recognition method was proposed based on the local binary patterns (LBP) and sub-image based feature sampling. The human face images were preprocessed by LBP technology to obtain LBP texture images. The sub-image division was performed on LBP texture images. The random feature sampling was completed based on sub-images in combination with random subspace method, and the based classifiers were trained by the randomly sampled features to build an ensemble of base classifiers. The multi-scale analysis was utilized to perform layered random sampling and divide multiple random samplings into layers with different sampling rate for each layer. The feature subsets with different dimensionalities were obtained with reduced total numbers of features, and the variety among based classifiers can be enhanced by cascade ensemble technology. The experiments were completed on the four face databases of ORL, Yale, Extended Yale B and CMU PIE. The results show that the proposed human face recognition method can achieve high runtime efficiency with high accuracy.
王进, 颉小凤, 胡明星, 邓欣, 陈乔松. 基于LBP预处理和子图像特征采样的人脸识别[J]. 江苏大学学报（自然科学版）, 2016, 37(1): 85-91.
WANG Jin, JIE Xiao-Feng, HU Ming-Xing, DENG Xin, CHEN Qiao-Song. Face recognition based on LBP preprocessing and sub-image feature sampling[J]. Journal of Jiangsu University(Natural Science Eidtion)
, 2016, 37(1): 85-91.
2016, Vol. 37 
